ACPI: thinkpad-acpi: keep track of module state
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / xfs / xfs_vnodeops.c
blob603459229904e0b6701b4a883b262041967e9a0a
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_types.h"
22 #include "xfs_bit.h"
23 #include "xfs_log.h"
24 #include "xfs_inum.h"
25 #include "xfs_trans.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
44 #include "xfs_bmap.h"
45 #include "xfs_attr.h"
46 #include "xfs_rw.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_utils.h"
50 #include "xfs_rtalloc.h"
51 #include "xfs_refcache.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
56 STATIC int
57 xfs_open(
58 bhv_desc_t *bdp,
59 cred_t *credp)
61 int mode;
62 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
63 xfs_inode_t *ip = XFS_BHVTOI(bdp);
65 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
66 return XFS_ERROR(EIO);
69 * If it's a directory with any blocks, read-ahead block 0
70 * as we're almost certain to have the next operation be a read there.
72 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
73 mode = xfs_ilock_map_shared(ip);
74 if (ip->i_d.di_nextents > 0)
75 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
76 xfs_iunlock(ip, mode);
78 return 0;
82 * xfs_getattr
84 STATIC int
85 xfs_getattr(
86 bhv_desc_t *bdp,
87 bhv_vattr_t *vap,
88 int flags,
89 cred_t *credp)
91 xfs_inode_t *ip;
92 xfs_mount_t *mp;
93 bhv_vnode_t *vp;
95 vp = BHV_TO_VNODE(bdp);
96 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
98 ip = XFS_BHVTOI(bdp);
99 mp = ip->i_mount;
101 if (XFS_FORCED_SHUTDOWN(mp))
102 return XFS_ERROR(EIO);
104 if (!(flags & ATTR_LAZY))
105 xfs_ilock(ip, XFS_ILOCK_SHARED);
107 vap->va_size = XFS_ISIZE(ip);
108 if (vap->va_mask == XFS_AT_SIZE)
109 goto all_done;
111 vap->va_nblocks =
112 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
113 vap->va_nodeid = ip->i_ino;
114 #if XFS_BIG_INUMS
115 vap->va_nodeid += mp->m_inoadd;
116 #endif
117 vap->va_nlink = ip->i_d.di_nlink;
120 * Quick exit for non-stat callers
122 if ((vap->va_mask &
123 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
124 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
125 goto all_done;
128 * Copy from in-core inode.
130 vap->va_mode = ip->i_d.di_mode;
131 vap->va_uid = ip->i_d.di_uid;
132 vap->va_gid = ip->i_d.di_gid;
133 vap->va_projid = ip->i_d.di_projid;
136 * Check vnode type block/char vs. everything else.
138 switch (ip->i_d.di_mode & S_IFMT) {
139 case S_IFBLK:
140 case S_IFCHR:
141 vap->va_rdev = ip->i_df.if_u2.if_rdev;
142 vap->va_blocksize = BLKDEV_IOSIZE;
143 break;
144 default:
145 vap->va_rdev = 0;
147 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
148 vap->va_blocksize = xfs_preferred_iosize(mp);
149 } else {
152 * If the file blocks are being allocated from a
153 * realtime partition, then return the inode's
154 * realtime extent size or the realtime volume's
155 * extent size.
157 vap->va_blocksize =
158 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
160 break;
163 vn_atime_to_timespec(vp, &vap->va_atime);
164 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
165 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
166 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
167 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
170 * Exit for stat callers. See if any of the rest of the fields
171 * to be filled in are needed.
173 if ((vap->va_mask &
174 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
175 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
176 goto all_done;
179 * Convert di_flags to xflags.
181 vap->va_xflags = xfs_ip2xflags(ip);
184 * Exit for inode revalidate. See if any of the rest of
185 * the fields to be filled in are needed.
187 if ((vap->va_mask &
188 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
189 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
190 goto all_done;
192 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
193 vap->va_nextents =
194 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
195 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
196 ip->i_d.di_nextents;
197 if (ip->i_afp)
198 vap->va_anextents =
199 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
200 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
201 ip->i_d.di_anextents;
202 else
203 vap->va_anextents = 0;
204 vap->va_gen = ip->i_d.di_gen;
206 all_done:
207 if (!(flags & ATTR_LAZY))
208 xfs_iunlock(ip, XFS_ILOCK_SHARED);
209 return 0;
214 * xfs_setattr
217 xfs_setattr(
218 bhv_desc_t *bdp,
219 bhv_vattr_t *vap,
220 int flags,
221 cred_t *credp)
223 xfs_inode_t *ip;
224 xfs_trans_t *tp;
225 xfs_mount_t *mp;
226 int mask;
227 int code;
228 uint lock_flags;
229 uint commit_flags=0;
230 uid_t uid=0, iuid=0;
231 gid_t gid=0, igid=0;
232 int timeflags = 0;
233 bhv_vnode_t *vp;
234 xfs_prid_t projid=0, iprojid=0;
235 int mandlock_before, mandlock_after;
236 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
237 int file_owner;
238 int need_iolock = 1;
240 vp = BHV_TO_VNODE(bdp);
241 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
243 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
244 return XFS_ERROR(EROFS);
247 * Cannot set certain attributes.
249 mask = vap->va_mask;
250 if (mask & XFS_AT_NOSET) {
251 return XFS_ERROR(EINVAL);
254 ip = XFS_BHVTOI(bdp);
255 mp = ip->i_mount;
257 if (XFS_FORCED_SHUTDOWN(mp))
258 return XFS_ERROR(EIO);
261 * Timestamps do not need to be logged and hence do not
262 * need to be done within a transaction.
264 if (mask & XFS_AT_UPDTIMES) {
265 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
266 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
267 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
268 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
269 xfs_ichgtime(ip, timeflags);
270 return 0;
273 olddquot1 = olddquot2 = NULL;
274 udqp = gdqp = NULL;
277 * If disk quotas is on, we make sure that the dquots do exist on disk,
278 * before we start any other transactions. Trying to do this later
279 * is messy. We don't care to take a readlock to look at the ids
280 * in inode here, because we can't hold it across the trans_reserve.
281 * If the IDs do change before we take the ilock, we're covered
282 * because the i_*dquot fields will get updated anyway.
284 if (XFS_IS_QUOTA_ON(mp) &&
285 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
286 uint qflags = 0;
288 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
289 uid = vap->va_uid;
290 qflags |= XFS_QMOPT_UQUOTA;
291 } else {
292 uid = ip->i_d.di_uid;
294 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
295 gid = vap->va_gid;
296 qflags |= XFS_QMOPT_GQUOTA;
297 } else {
298 gid = ip->i_d.di_gid;
300 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
301 projid = vap->va_projid;
302 qflags |= XFS_QMOPT_PQUOTA;
303 } else {
304 projid = ip->i_d.di_projid;
307 * We take a reference when we initialize udqp and gdqp,
308 * so it is important that we never blindly double trip on
309 * the same variable. See xfs_create() for an example.
311 ASSERT(udqp == NULL);
312 ASSERT(gdqp == NULL);
313 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
314 &udqp, &gdqp);
315 if (code)
316 return code;
320 * For the other attributes, we acquire the inode lock and
321 * first do an error checking pass.
323 tp = NULL;
324 lock_flags = XFS_ILOCK_EXCL;
325 if (flags & ATTR_NOLOCK)
326 need_iolock = 0;
327 if (!(mask & XFS_AT_SIZE)) {
328 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
329 (mp->m_flags & XFS_MOUNT_WSYNC)) {
330 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
331 commit_flags = 0;
332 if ((code = xfs_trans_reserve(tp, 0,
333 XFS_ICHANGE_LOG_RES(mp), 0,
334 0, 0))) {
335 lock_flags = 0;
336 goto error_return;
339 } else {
340 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
341 !(flags & ATTR_DMI)) {
342 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
343 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
344 vap->va_size, 0, dmflags, NULL);
345 if (code) {
346 lock_flags = 0;
347 goto error_return;
350 if (need_iolock)
351 lock_flags |= XFS_IOLOCK_EXCL;
354 xfs_ilock(ip, lock_flags);
356 /* boolean: are we the file owner? */
357 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
360 * Change various properties of a file.
361 * Only the owner or users with CAP_FOWNER
362 * capability may do these things.
364 if (mask &
365 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
366 XFS_AT_GID|XFS_AT_PROJID)) {
368 * CAP_FOWNER overrides the following restrictions:
370 * The user ID of the calling process must be equal
371 * to the file owner ID, except in cases where the
372 * CAP_FSETID capability is applicable.
374 if (!file_owner && !capable(CAP_FOWNER)) {
375 code = XFS_ERROR(EPERM);
376 goto error_return;
380 * CAP_FSETID overrides the following restrictions:
382 * The effective user ID of the calling process shall match
383 * the file owner when setting the set-user-ID and
384 * set-group-ID bits on that file.
386 * The effective group ID or one of the supplementary group
387 * IDs of the calling process shall match the group owner of
388 * the file when setting the set-group-ID bit on that file
390 if (mask & XFS_AT_MODE) {
391 mode_t m = 0;
393 if ((vap->va_mode & S_ISUID) && !file_owner)
394 m |= S_ISUID;
395 if ((vap->va_mode & S_ISGID) &&
396 !in_group_p((gid_t)ip->i_d.di_gid))
397 m |= S_ISGID;
398 #if 0
399 /* Linux allows this, Irix doesn't. */
400 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
401 m |= S_ISVTX;
402 #endif
403 if (m && !capable(CAP_FSETID))
404 vap->va_mode &= ~m;
409 * Change file ownership. Must be the owner or privileged.
410 * If the system was configured with the "restricted_chown"
411 * option, the owner is not permitted to give away the file,
412 * and can change the group id only to a group of which he
413 * or she is a member.
415 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
417 * These IDs could have changed since we last looked at them.
418 * But, we're assured that if the ownership did change
419 * while we didn't have the inode locked, inode's dquot(s)
420 * would have changed also.
422 iuid = ip->i_d.di_uid;
423 iprojid = ip->i_d.di_projid;
424 igid = ip->i_d.di_gid;
425 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
426 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
427 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
428 iprojid;
431 * CAP_CHOWN overrides the following restrictions:
433 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
434 * shall override the restriction that a process cannot
435 * change the user ID of a file it owns and the restriction
436 * that the group ID supplied to the chown() function
437 * shall be equal to either the group ID or one of the
438 * supplementary group IDs of the calling process.
440 if (restricted_chown &&
441 (iuid != uid || (igid != gid &&
442 !in_group_p((gid_t)gid))) &&
443 !capable(CAP_CHOWN)) {
444 code = XFS_ERROR(EPERM);
445 goto error_return;
448 * Do a quota reservation only if uid/projid/gid is actually
449 * going to change.
451 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
452 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
453 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
454 ASSERT(tp);
455 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
456 capable(CAP_FOWNER) ?
457 XFS_QMOPT_FORCE_RES : 0);
458 if (code) /* out of quota */
459 goto error_return;
464 * Truncate file. Must have write permission and not be a directory.
466 if (mask & XFS_AT_SIZE) {
467 /* Short circuit the truncate case for zero length files */
468 if ((vap->va_size == 0) &&
469 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
470 xfs_iunlock(ip, XFS_ILOCK_EXCL);
471 lock_flags &= ~XFS_ILOCK_EXCL;
472 if (mask & XFS_AT_CTIME)
473 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
474 code = 0;
475 goto error_return;
478 if (VN_ISDIR(vp)) {
479 code = XFS_ERROR(EISDIR);
480 goto error_return;
481 } else if (!VN_ISREG(vp)) {
482 code = XFS_ERROR(EINVAL);
483 goto error_return;
486 * Make sure that the dquots are attached to the inode.
488 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
489 goto error_return;
493 * Change file access or modified times.
495 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
496 if (!file_owner) {
497 if ((flags & ATTR_UTIME) &&
498 !capable(CAP_FOWNER)) {
499 code = XFS_ERROR(EPERM);
500 goto error_return;
506 * Change extent size or realtime flag.
508 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
510 * Can't change extent size if any extents are allocated.
512 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
513 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
514 vap->va_extsize) ) {
515 code = XFS_ERROR(EINVAL); /* EFBIG? */
516 goto error_return;
520 * Can't change realtime flag if any extents are allocated.
522 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
523 (mask & XFS_AT_XFLAGS) &&
524 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
525 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
526 code = XFS_ERROR(EINVAL); /* EFBIG? */
527 goto error_return;
530 * Extent size must be a multiple of the appropriate block
531 * size, if set at all.
533 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
534 xfs_extlen_t size;
536 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
537 ((mask & XFS_AT_XFLAGS) &&
538 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
539 size = mp->m_sb.sb_rextsize <<
540 mp->m_sb.sb_blocklog;
541 } else {
542 size = mp->m_sb.sb_blocksize;
544 if (vap->va_extsize % size) {
545 code = XFS_ERROR(EINVAL);
546 goto error_return;
550 * If realtime flag is set then must have realtime data.
552 if ((mask & XFS_AT_XFLAGS) &&
553 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
554 if ((mp->m_sb.sb_rblocks == 0) ||
555 (mp->m_sb.sb_rextsize == 0) ||
556 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
557 code = XFS_ERROR(EINVAL);
558 goto error_return;
563 * Can't modify an immutable/append-only file unless
564 * we have appropriate permission.
566 if ((mask & XFS_AT_XFLAGS) &&
567 (ip->i_d.di_flags &
568 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
569 (vap->va_xflags &
570 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
571 !capable(CAP_LINUX_IMMUTABLE)) {
572 code = XFS_ERROR(EPERM);
573 goto error_return;
578 * Now we can make the changes. Before we join the inode
579 * to the transaction, if XFS_AT_SIZE is set then take care of
580 * the part of the truncation that must be done without the
581 * inode lock. This needs to be done before joining the inode
582 * to the transaction, because the inode cannot be unlocked
583 * once it is a part of the transaction.
585 if (mask & XFS_AT_SIZE) {
586 code = 0;
587 if ((vap->va_size > ip->i_size) &&
588 (flags & ATTR_NOSIZETOK) == 0) {
589 code = xfs_igrow_start(ip, vap->va_size, credp);
591 xfs_iunlock(ip, XFS_ILOCK_EXCL);
594 * We are going to log the inode size change in this
595 * transaction so any previous writes that are beyond the on
596 * disk EOF and the new EOF that have not been written out need
597 * to be written here. If we do not write the data out, we
598 * expose ourselves to the null files problem.
600 * Only flush from the on disk size to the smaller of the in
601 * memory file size or the new size as that's the range we
602 * really care about here and prevents waiting for other data
603 * not within the range we care about here.
605 if (!code &&
606 (ip->i_size != ip->i_d.di_size) &&
607 (vap->va_size > ip->i_d.di_size)) {
608 code = bhv_vop_flush_pages(XFS_ITOV(ip),
609 ip->i_d.di_size, vap->va_size,
610 XFS_B_ASYNC, FI_NONE);
613 /* wait for all I/O to complete */
614 vn_iowait(vp);
616 if (!code)
617 code = xfs_itruncate_data(ip, vap->va_size);
618 if (code) {
619 ASSERT(tp == NULL);
620 lock_flags &= ~XFS_ILOCK_EXCL;
621 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
622 goto error_return;
624 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
625 if ((code = xfs_trans_reserve(tp, 0,
626 XFS_ITRUNCATE_LOG_RES(mp), 0,
627 XFS_TRANS_PERM_LOG_RES,
628 XFS_ITRUNCATE_LOG_COUNT))) {
629 xfs_trans_cancel(tp, 0);
630 if (need_iolock)
631 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
632 return code;
634 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
635 xfs_ilock(ip, XFS_ILOCK_EXCL);
638 if (tp) {
639 xfs_trans_ijoin(tp, ip, lock_flags);
640 xfs_trans_ihold(tp, ip);
643 /* determine whether mandatory locking mode changes */
644 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
647 * Truncate file. Must have write permission and not be a directory.
649 if (mask & XFS_AT_SIZE) {
650 if (vap->va_size > ip->i_size) {
651 xfs_igrow_finish(tp, ip, vap->va_size,
652 !(flags & ATTR_DMI));
653 } else if ((vap->va_size <= ip->i_size) ||
654 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
656 * signal a sync transaction unless
657 * we're truncating an already unlinked
658 * file on a wsync filesystem
660 code = xfs_itruncate_finish(&tp, ip,
661 (xfs_fsize_t)vap->va_size,
662 XFS_DATA_FORK,
663 ((ip->i_d.di_nlink != 0 ||
664 !(mp->m_flags & XFS_MOUNT_WSYNC))
665 ? 1 : 0));
666 if (code)
667 goto abort_return;
669 * Truncated "down", so we're removing references
670 * to old data here - if we now delay flushing for
671 * a long time, we expose ourselves unduly to the
672 * notorious NULL files problem. So, we mark this
673 * vnode and flush it when the file is closed, and
674 * do not wait the usual (long) time for writeout.
676 VTRUNCATE(vp);
679 * Have to do this even if the file's size doesn't change.
681 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
685 * Change file access modes.
687 if (mask & XFS_AT_MODE) {
688 ip->i_d.di_mode &= S_IFMT;
689 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
691 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
692 timeflags |= XFS_ICHGTIME_CHG;
696 * Change file ownership. Must be the owner or privileged.
697 * If the system was configured with the "restricted_chown"
698 * option, the owner is not permitted to give away the file,
699 * and can change the group id only to a group of which he
700 * or she is a member.
702 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
704 * CAP_FSETID overrides the following restrictions:
706 * The set-user-ID and set-group-ID bits of a file will be
707 * cleared upon successful return from chown()
709 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
710 !capable(CAP_FSETID)) {
711 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
715 * Change the ownerships and register quota modifications
716 * in the transaction.
718 if (iuid != uid) {
719 if (XFS_IS_UQUOTA_ON(mp)) {
720 ASSERT(mask & XFS_AT_UID);
721 ASSERT(udqp);
722 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
723 &ip->i_udquot, udqp);
725 ip->i_d.di_uid = uid;
727 if (igid != gid) {
728 if (XFS_IS_GQUOTA_ON(mp)) {
729 ASSERT(!XFS_IS_PQUOTA_ON(mp));
730 ASSERT(mask & XFS_AT_GID);
731 ASSERT(gdqp);
732 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
733 &ip->i_gdquot, gdqp);
735 ip->i_d.di_gid = gid;
737 if (iprojid != projid) {
738 if (XFS_IS_PQUOTA_ON(mp)) {
739 ASSERT(!XFS_IS_GQUOTA_ON(mp));
740 ASSERT(mask & XFS_AT_PROJID);
741 ASSERT(gdqp);
742 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
743 &ip->i_gdquot, gdqp);
745 ip->i_d.di_projid = projid;
747 * We may have to rev the inode as well as
748 * the superblock version number since projids didn't
749 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
751 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
752 xfs_bump_ino_vers2(tp, ip);
755 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
756 timeflags |= XFS_ICHGTIME_CHG;
761 * Change file access or modified times.
763 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
764 if (mask & XFS_AT_ATIME) {
765 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
766 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
767 ip->i_update_core = 1;
768 timeflags &= ~XFS_ICHGTIME_ACC;
770 if (mask & XFS_AT_MTIME) {
771 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
772 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
773 timeflags &= ~XFS_ICHGTIME_MOD;
774 timeflags |= XFS_ICHGTIME_CHG;
776 if (tp && (flags & ATTR_UTIME))
777 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
781 * Change XFS-added attributes.
783 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
784 if (mask & XFS_AT_EXTSIZE) {
786 * Converting bytes to fs blocks.
788 ip->i_d.di_extsize = vap->va_extsize >>
789 mp->m_sb.sb_blocklog;
791 if (mask & XFS_AT_XFLAGS) {
792 uint di_flags;
794 /* can't set PREALLOC this way, just preserve it */
795 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
796 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
797 di_flags |= XFS_DIFLAG_IMMUTABLE;
798 if (vap->va_xflags & XFS_XFLAG_APPEND)
799 di_flags |= XFS_DIFLAG_APPEND;
800 if (vap->va_xflags & XFS_XFLAG_SYNC)
801 di_flags |= XFS_DIFLAG_SYNC;
802 if (vap->va_xflags & XFS_XFLAG_NOATIME)
803 di_flags |= XFS_DIFLAG_NOATIME;
804 if (vap->va_xflags & XFS_XFLAG_NODUMP)
805 di_flags |= XFS_DIFLAG_NODUMP;
806 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
807 di_flags |= XFS_DIFLAG_PROJINHERIT;
808 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
809 di_flags |= XFS_DIFLAG_NODEFRAG;
810 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
811 di_flags |= XFS_DIFLAG_FILESTREAM;
812 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
813 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
814 di_flags |= XFS_DIFLAG_RTINHERIT;
815 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
816 di_flags |= XFS_DIFLAG_NOSYMLINKS;
817 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
818 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
819 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
820 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
821 di_flags |= XFS_DIFLAG_REALTIME;
822 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
823 } else {
824 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
826 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
827 di_flags |= XFS_DIFLAG_EXTSIZE;
829 ip->i_d.di_flags = di_flags;
831 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
832 timeflags |= XFS_ICHGTIME_CHG;
836 * Change file inode change time only if XFS_AT_CTIME set
837 * AND we have been called by a DMI function.
840 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
841 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
842 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
843 ip->i_update_core = 1;
844 timeflags &= ~XFS_ICHGTIME_CHG;
848 * Send out timestamp changes that need to be set to the
849 * current time. Not done when called by a DMI function.
851 if (timeflags && !(flags & ATTR_DMI))
852 xfs_ichgtime(ip, timeflags);
854 XFS_STATS_INC(xs_ig_attrchg);
857 * If this is a synchronous mount, make sure that the
858 * transaction goes to disk before returning to the user.
859 * This is slightly sub-optimal in that truncates require
860 * two sync transactions instead of one for wsync filesystems.
861 * One for the truncate and one for the timestamps since we
862 * don't want to change the timestamps unless we're sure the
863 * truncate worked. Truncates are less than 1% of the laddis
864 * mix so this probably isn't worth the trouble to optimize.
866 code = 0;
867 if (tp) {
868 if (mp->m_flags & XFS_MOUNT_WSYNC)
869 xfs_trans_set_sync(tp);
871 code = xfs_trans_commit(tp, commit_flags);
875 * If the (regular) file's mandatory locking mode changed, then
876 * notify the vnode. We do this under the inode lock to prevent
877 * racing calls to vop_vnode_change.
879 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
880 if (mandlock_before != mandlock_after) {
881 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING,
882 mandlock_after);
885 xfs_iunlock(ip, lock_flags);
888 * Release any dquot(s) the inode had kept before chown.
890 XFS_QM_DQRELE(mp, olddquot1);
891 XFS_QM_DQRELE(mp, olddquot2);
892 XFS_QM_DQRELE(mp, udqp);
893 XFS_QM_DQRELE(mp, gdqp);
895 if (code) {
896 return code;
899 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
900 !(flags & ATTR_DMI)) {
901 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
902 NULL, DM_RIGHT_NULL, NULL, NULL,
903 0, 0, AT_DELAY_FLAG(flags));
905 return 0;
907 abort_return:
908 commit_flags |= XFS_TRANS_ABORT;
909 /* FALLTHROUGH */
910 error_return:
911 XFS_QM_DQRELE(mp, udqp);
912 XFS_QM_DQRELE(mp, gdqp);
913 if (tp) {
914 xfs_trans_cancel(tp, commit_flags);
916 if (lock_flags != 0) {
917 xfs_iunlock(ip, lock_flags);
919 return code;
924 * xfs_access
925 * Null conversion from vnode mode bits to inode mode bits, as in efs.
927 STATIC int
928 xfs_access(
929 bhv_desc_t *bdp,
930 int mode,
931 cred_t *credp)
933 xfs_inode_t *ip;
934 int error;
936 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
937 (inst_t *)__return_address);
939 ip = XFS_BHVTOI(bdp);
940 xfs_ilock(ip, XFS_ILOCK_SHARED);
941 error = xfs_iaccess(ip, mode, credp);
942 xfs_iunlock(ip, XFS_ILOCK_SHARED);
943 return error;
948 * The maximum pathlen is 1024 bytes. Since the minimum file system
949 * blocksize is 512 bytes, we can get a max of 2 extents back from
950 * bmapi.
952 #define SYMLINK_MAPS 2
955 * xfs_readlink
958 STATIC int
959 xfs_readlink(
960 bhv_desc_t *bdp,
961 uio_t *uiop,
962 int ioflags,
963 cred_t *credp)
965 xfs_inode_t *ip;
966 int count;
967 xfs_off_t offset;
968 int pathlen;
969 bhv_vnode_t *vp;
970 int error = 0;
971 xfs_mount_t *mp;
972 int nmaps;
973 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
974 xfs_daddr_t d;
975 int byte_cnt;
976 int n;
977 xfs_buf_t *bp;
979 vp = BHV_TO_VNODE(bdp);
980 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
982 ip = XFS_BHVTOI(bdp);
983 mp = ip->i_mount;
985 if (XFS_FORCED_SHUTDOWN(mp))
986 return XFS_ERROR(EIO);
988 xfs_ilock(ip, XFS_ILOCK_SHARED);
990 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
992 offset = uiop->uio_offset;
993 count = uiop->uio_resid;
995 if (offset < 0) {
996 error = XFS_ERROR(EINVAL);
997 goto error_return;
999 if (count <= 0) {
1000 error = 0;
1001 goto error_return;
1005 * See if the symlink is stored inline.
1007 pathlen = (int)ip->i_d.di_size;
1009 if (ip->i_df.if_flags & XFS_IFINLINE) {
1010 error = xfs_uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1012 else {
1014 * Symlink not inline. Call bmap to get it in.
1016 nmaps = SYMLINK_MAPS;
1018 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1019 0, NULL, 0, mval, &nmaps, NULL, NULL);
1021 if (error) {
1022 goto error_return;
1025 for (n = 0; n < nmaps; n++) {
1026 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1027 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1028 bp = xfs_buf_read(mp->m_ddev_targp, d,
1029 BTOBB(byte_cnt), 0);
1030 error = XFS_BUF_GETERROR(bp);
1031 if (error) {
1032 xfs_ioerror_alert("xfs_readlink",
1033 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1034 xfs_buf_relse(bp);
1035 goto error_return;
1037 if (pathlen < byte_cnt)
1038 byte_cnt = pathlen;
1039 pathlen -= byte_cnt;
1041 error = xfs_uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1042 xfs_buf_relse (bp);
1047 error_return:
1048 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1049 return error;
1054 * xfs_fsync
1056 * This is called to sync the inode and its data out to disk.
1057 * We need to hold the I/O lock while flushing the data, and
1058 * the inode lock while flushing the inode. The inode lock CANNOT
1059 * be held while flushing the data, so acquire after we're done
1060 * with that.
1062 STATIC int
1063 xfs_fsync(
1064 bhv_desc_t *bdp,
1065 int flag,
1066 cred_t *credp,
1067 xfs_off_t start,
1068 xfs_off_t stop)
1070 xfs_inode_t *ip;
1071 xfs_trans_t *tp;
1072 int error;
1073 int log_flushed = 0, changed = 1;
1075 vn_trace_entry(BHV_TO_VNODE(bdp),
1076 __FUNCTION__, (inst_t *)__return_address);
1078 ip = XFS_BHVTOI(bdp);
1080 ASSERT(start >= 0 && stop >= -1);
1082 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1083 return XFS_ERROR(EIO);
1085 if (flag & FSYNC_DATA)
1086 filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping);
1089 * We always need to make sure that the required inode state
1090 * is safe on disk. The vnode might be clean but because
1091 * of committed transactions that haven't hit the disk yet.
1092 * Likewise, there could be unflushed non-transactional
1093 * changes to the inode core that have to go to disk.
1095 * The following code depends on one assumption: that
1096 * any transaction that changes an inode logs the core
1097 * because it has to change some field in the inode core
1098 * (typically nextents or nblocks). That assumption
1099 * implies that any transactions against an inode will
1100 * catch any non-transactional updates. If inode-altering
1101 * transactions exist that violate this assumption, the
1102 * code breaks. Right now, it figures that if the involved
1103 * update_* field is clear and the inode is unpinned, the
1104 * inode is clean. Either it's been flushed or it's been
1105 * committed and the commit has hit the disk unpinning the inode.
1106 * (Note that xfs_inode_item_format() called at commit clears
1107 * the update_* fields.)
1109 xfs_ilock(ip, XFS_ILOCK_SHARED);
1111 /* If we are flushing data then we care about update_size
1112 * being set, otherwise we care about update_core
1114 if ((flag & FSYNC_DATA) ?
1115 (ip->i_update_size == 0) :
1116 (ip->i_update_core == 0)) {
1118 * Timestamps/size haven't changed since last inode
1119 * flush or inode transaction commit. That means
1120 * either nothing got written or a transaction
1121 * committed which caught the updates. If the
1122 * latter happened and the transaction hasn't
1123 * hit the disk yet, the inode will be still
1124 * be pinned. If it is, force the log.
1127 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1129 if (xfs_ipincount(ip)) {
1130 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1131 XFS_LOG_FORCE |
1132 ((flag & FSYNC_WAIT)
1133 ? XFS_LOG_SYNC : 0),
1134 &log_flushed);
1135 } else {
1137 * If the inode is not pinned and nothing
1138 * has changed we don't need to flush the
1139 * cache.
1141 changed = 0;
1143 error = 0;
1144 } else {
1146 * Kick off a transaction to log the inode
1147 * core to get the updates. Make it
1148 * sync if FSYNC_WAIT is passed in (which
1149 * is done by everybody but specfs). The
1150 * sync transaction will also force the log.
1152 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1153 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1154 if ((error = xfs_trans_reserve(tp, 0,
1155 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1156 0, 0, 0))) {
1157 xfs_trans_cancel(tp, 0);
1158 return error;
1160 xfs_ilock(ip, XFS_ILOCK_EXCL);
1163 * Note - it's possible that we might have pushed
1164 * ourselves out of the way during trans_reserve
1165 * which would flush the inode. But there's no
1166 * guarantee that the inode buffer has actually
1167 * gone out yet (it's delwri). Plus the buffer
1168 * could be pinned anyway if it's part of an
1169 * inode in another recent transaction. So we
1170 * play it safe and fire off the transaction anyway.
1172 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1173 xfs_trans_ihold(tp, ip);
1174 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1175 if (flag & FSYNC_WAIT)
1176 xfs_trans_set_sync(tp);
1177 error = _xfs_trans_commit(tp, 0, &log_flushed);
1179 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1182 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1184 * If the log write didn't issue an ordered tag we need
1185 * to flush the disk cache for the data device now.
1187 if (!log_flushed)
1188 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1191 * If this inode is on the RT dev we need to flush that
1192 * cache as well.
1194 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1195 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1198 return error;
1202 * This is called by xfs_inactive to free any blocks beyond eof
1203 * when the link count isn't zero and by xfs_dm_punch_hole() when
1204 * punching a hole to EOF.
1207 xfs_free_eofblocks(
1208 xfs_mount_t *mp,
1209 xfs_inode_t *ip,
1210 int flags)
1212 xfs_trans_t *tp;
1213 int error;
1214 xfs_fileoff_t end_fsb;
1215 xfs_fileoff_t last_fsb;
1216 xfs_filblks_t map_len;
1217 int nimaps;
1218 xfs_bmbt_irec_t imap;
1219 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
1222 * Figure out if there are any blocks beyond the end
1223 * of the file. If not, then there is nothing to do.
1225 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
1226 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1227 map_len = last_fsb - end_fsb;
1228 if (map_len <= 0)
1229 return 0;
1231 nimaps = 1;
1232 xfs_ilock(ip, XFS_ILOCK_SHARED);
1233 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1234 NULL, 0, &imap, &nimaps, NULL, NULL);
1235 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1237 if (!error && (nimaps != 0) &&
1238 (imap.br_startblock != HOLESTARTBLOCK ||
1239 ip->i_delayed_blks)) {
1241 * Attach the dquots to the inode up front.
1243 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1244 return error;
1247 * There are blocks after the end of file.
1248 * Free them up now by truncating the file to
1249 * its current size.
1251 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1254 * Do the xfs_itruncate_start() call before
1255 * reserving any log space because
1256 * itruncate_start will call into the buffer
1257 * cache and we can't
1258 * do that within a transaction.
1260 if (use_iolock)
1261 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1262 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1263 ip->i_size);
1264 if (error) {
1265 xfs_trans_cancel(tp, 0);
1266 if (use_iolock)
1267 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1268 return error;
1271 error = xfs_trans_reserve(tp, 0,
1272 XFS_ITRUNCATE_LOG_RES(mp),
1273 0, XFS_TRANS_PERM_LOG_RES,
1274 XFS_ITRUNCATE_LOG_COUNT);
1275 if (error) {
1276 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1277 xfs_trans_cancel(tp, 0);
1278 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1279 return error;
1282 xfs_ilock(ip, XFS_ILOCK_EXCL);
1283 xfs_trans_ijoin(tp, ip,
1284 XFS_IOLOCK_EXCL |
1285 XFS_ILOCK_EXCL);
1286 xfs_trans_ihold(tp, ip);
1288 error = xfs_itruncate_finish(&tp, ip,
1289 ip->i_size,
1290 XFS_DATA_FORK,
1293 * If we get an error at this point we
1294 * simply don't bother truncating the file.
1296 if (error) {
1297 xfs_trans_cancel(tp,
1298 (XFS_TRANS_RELEASE_LOG_RES |
1299 XFS_TRANS_ABORT));
1300 } else {
1301 error = xfs_trans_commit(tp,
1302 XFS_TRANS_RELEASE_LOG_RES);
1304 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
1305 : XFS_ILOCK_EXCL));
1307 return error;
1311 * Free a symlink that has blocks associated with it.
1313 STATIC int
1314 xfs_inactive_symlink_rmt(
1315 xfs_inode_t *ip,
1316 xfs_trans_t **tpp)
1318 xfs_buf_t *bp;
1319 int committed;
1320 int done;
1321 int error;
1322 xfs_fsblock_t first_block;
1323 xfs_bmap_free_t free_list;
1324 int i;
1325 xfs_mount_t *mp;
1326 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1327 int nmaps;
1328 xfs_trans_t *ntp;
1329 int size;
1330 xfs_trans_t *tp;
1332 tp = *tpp;
1333 mp = ip->i_mount;
1334 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1336 * We're freeing a symlink that has some
1337 * blocks allocated to it. Free the
1338 * blocks here. We know that we've got
1339 * either 1 or 2 extents and that we can
1340 * free them all in one bunmapi call.
1342 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1343 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1344 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1345 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1346 xfs_trans_cancel(tp, 0);
1347 *tpp = NULL;
1348 return error;
1351 * Lock the inode, fix the size, and join it to the transaction.
1352 * Hold it so in the normal path, we still have it locked for
1353 * the second transaction. In the error paths we need it
1354 * held so the cancel won't rele it, see below.
1356 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1357 size = (int)ip->i_d.di_size;
1358 ip->i_d.di_size = 0;
1359 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1360 xfs_trans_ihold(tp, ip);
1361 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1363 * Find the block(s) so we can inval and unmap them.
1365 done = 0;
1366 XFS_BMAP_INIT(&free_list, &first_block);
1367 nmaps = ARRAY_SIZE(mval);
1368 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1369 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1370 &free_list, NULL)))
1371 goto error0;
1373 * Invalidate the block(s).
1375 for (i = 0; i < nmaps; i++) {
1376 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1377 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1378 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1379 xfs_trans_binval(tp, bp);
1382 * Unmap the dead block(s) to the free_list.
1384 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1385 &first_block, &free_list, NULL, &done)))
1386 goto error1;
1387 ASSERT(done);
1389 * Commit the first transaction. This logs the EFI and the inode.
1391 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1392 goto error1;
1394 * The transaction must have been committed, since there were
1395 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1396 * The new tp has the extent freeing and EFDs.
1398 ASSERT(committed);
1400 * The first xact was committed, so add the inode to the new one.
1401 * Mark it dirty so it will be logged and moved forward in the log as
1402 * part of every commit.
1404 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1405 xfs_trans_ihold(tp, ip);
1406 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1408 * Get a new, empty transaction to return to our caller.
1410 ntp = xfs_trans_dup(tp);
1412 * Commit the transaction containing extent freeing and EFDs.
1413 * If we get an error on the commit here or on the reserve below,
1414 * we need to unlock the inode since the new transaction doesn't
1415 * have the inode attached.
1417 error = xfs_trans_commit(tp, 0);
1418 tp = ntp;
1419 if (error) {
1420 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1421 goto error0;
1424 * Remove the memory for extent descriptions (just bookkeeping).
1426 if (ip->i_df.if_bytes)
1427 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1428 ASSERT(ip->i_df.if_bytes == 0);
1430 * Put an itruncate log reservation in the new transaction
1431 * for our caller.
1433 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1434 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1435 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1436 goto error0;
1439 * Return with the inode locked but not joined to the transaction.
1441 *tpp = tp;
1442 return 0;
1444 error1:
1445 xfs_bmap_cancel(&free_list);
1446 error0:
1448 * Have to come here with the inode locked and either
1449 * (held and in the transaction) or (not in the transaction).
1450 * If the inode isn't held then cancel would iput it, but
1451 * that's wrong since this is inactive and the vnode ref
1452 * count is 0 already.
1453 * Cancel won't do anything to the inode if held, but it still
1454 * needs to be locked until the cancel is done, if it was
1455 * joined to the transaction.
1457 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1458 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1459 *tpp = NULL;
1460 return error;
1464 STATIC int
1465 xfs_inactive_symlink_local(
1466 xfs_inode_t *ip,
1467 xfs_trans_t **tpp)
1469 int error;
1471 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1473 * We're freeing a symlink which fit into
1474 * the inode. Just free the memory used
1475 * to hold the old symlink.
1477 error = xfs_trans_reserve(*tpp, 0,
1478 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1479 0, XFS_TRANS_PERM_LOG_RES,
1480 XFS_ITRUNCATE_LOG_COUNT);
1482 if (error) {
1483 xfs_trans_cancel(*tpp, 0);
1484 *tpp = NULL;
1485 return error;
1487 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1490 * Zero length symlinks _can_ exist.
1492 if (ip->i_df.if_bytes > 0) {
1493 xfs_idata_realloc(ip,
1494 -(ip->i_df.if_bytes),
1495 XFS_DATA_FORK);
1496 ASSERT(ip->i_df.if_bytes == 0);
1498 return 0;
1501 STATIC int
1502 xfs_inactive_attrs(
1503 xfs_inode_t *ip,
1504 xfs_trans_t **tpp)
1506 xfs_trans_t *tp;
1507 int error;
1508 xfs_mount_t *mp;
1510 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1511 tp = *tpp;
1512 mp = ip->i_mount;
1513 ASSERT(ip->i_d.di_forkoff != 0);
1514 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1515 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1517 error = xfs_attr_inactive(ip);
1518 if (error) {
1519 *tpp = NULL;
1520 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1521 return error; /* goto out */
1524 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1525 error = xfs_trans_reserve(tp, 0,
1526 XFS_IFREE_LOG_RES(mp),
1527 0, XFS_TRANS_PERM_LOG_RES,
1528 XFS_INACTIVE_LOG_COUNT);
1529 if (error) {
1530 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1531 xfs_trans_cancel(tp, 0);
1532 *tpp = NULL;
1533 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1534 return error;
1537 xfs_ilock(ip, XFS_ILOCK_EXCL);
1538 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1539 xfs_trans_ihold(tp, ip);
1540 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1542 ASSERT(ip->i_d.di_anextents == 0);
1544 *tpp = tp;
1545 return 0;
1548 STATIC int
1549 xfs_release(
1550 bhv_desc_t *bdp)
1552 xfs_inode_t *ip;
1553 bhv_vnode_t *vp;
1554 xfs_mount_t *mp;
1555 int error;
1557 vp = BHV_TO_VNODE(bdp);
1558 ip = XFS_BHVTOI(bdp);
1559 mp = ip->i_mount;
1561 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
1562 return 0;
1564 /* If this is a read-only mount, don't do this (would generate I/O) */
1565 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1566 return 0;
1568 if (!XFS_FORCED_SHUTDOWN(mp)) {
1570 * If we are using filestreams, and we have an unlinked
1571 * file that we are processing the last close on, then nothing
1572 * will be able to reopen and write to this file. Purge this
1573 * inode from the filestreams cache so that it doesn't delay
1574 * teardown of the inode.
1576 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1577 xfs_filestream_deassociate(ip);
1580 * If we previously truncated this file and removed old data
1581 * in the process, we want to initiate "early" writeout on
1582 * the last close. This is an attempt to combat the notorious
1583 * NULL files problem which is particularly noticable from a
1584 * truncate down, buffered (re-)write (delalloc), followed by
1585 * a crash. What we are effectively doing here is
1586 * significantly reducing the time window where we'd otherwise
1587 * be exposed to that problem.
1589 if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
1590 bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
1593 #ifdef HAVE_REFCACHE
1594 /* If we are in the NFS reference cache then don't do this now */
1595 if (ip->i_refcache)
1596 return 0;
1597 #endif
1599 if (ip->i_d.di_nlink != 0) {
1600 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1601 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1602 ip->i_delayed_blks > 0)) &&
1603 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1604 (!(ip->i_d.di_flags &
1605 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1606 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1607 if (error)
1608 return error;
1609 /* Update linux inode block count after free above */
1610 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1611 ip->i_d.di_nblocks + ip->i_delayed_blks);
1615 return 0;
1619 * xfs_inactive
1621 * This is called when the vnode reference count for the vnode
1622 * goes to zero. If the file has been unlinked, then it must
1623 * now be truncated. Also, we clear all of the read-ahead state
1624 * kept for the inode here since the file is now closed.
1626 STATIC int
1627 xfs_inactive(
1628 bhv_desc_t *bdp,
1629 cred_t *credp)
1631 xfs_inode_t *ip;
1632 bhv_vnode_t *vp;
1633 xfs_bmap_free_t free_list;
1634 xfs_fsblock_t first_block;
1635 int committed;
1636 xfs_trans_t *tp;
1637 xfs_mount_t *mp;
1638 int error;
1639 int truncate;
1641 vp = BHV_TO_VNODE(bdp);
1642 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1644 ip = XFS_BHVTOI(bdp);
1647 * If the inode is already free, then there can be nothing
1648 * to clean up here.
1650 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1651 ASSERT(ip->i_df.if_real_bytes == 0);
1652 ASSERT(ip->i_df.if_broot_bytes == 0);
1653 return VN_INACTIVE_CACHE;
1657 * Only do a truncate if it's a regular file with
1658 * some actual space in it. It's OK to look at the
1659 * inode's fields without the lock because we're the
1660 * only one with a reference to the inode.
1662 truncate = ((ip->i_d.di_nlink == 0) &&
1663 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1664 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1665 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1667 mp = ip->i_mount;
1669 if (ip->i_d.di_nlink == 0 &&
1670 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1671 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1674 error = 0;
1676 /* If this is a read-only mount, don't do this (would generate I/O) */
1677 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1678 goto out;
1680 if (ip->i_d.di_nlink != 0) {
1681 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1682 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1683 ip->i_delayed_blks > 0)) &&
1684 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1685 (!(ip->i_d.di_flags &
1686 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1687 (ip->i_delayed_blks != 0)))) {
1688 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1689 if (error)
1690 return VN_INACTIVE_CACHE;
1691 /* Update linux inode block count after free above */
1692 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1693 ip->i_d.di_nblocks + ip->i_delayed_blks);
1695 goto out;
1698 ASSERT(ip->i_d.di_nlink == 0);
1700 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1701 return VN_INACTIVE_CACHE;
1703 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1704 if (truncate) {
1706 * Do the xfs_itruncate_start() call before
1707 * reserving any log space because itruncate_start
1708 * will call into the buffer cache and we can't
1709 * do that within a transaction.
1711 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1713 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1714 if (error) {
1715 xfs_trans_cancel(tp, 0);
1716 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1717 return VN_INACTIVE_CACHE;
1720 error = xfs_trans_reserve(tp, 0,
1721 XFS_ITRUNCATE_LOG_RES(mp),
1722 0, XFS_TRANS_PERM_LOG_RES,
1723 XFS_ITRUNCATE_LOG_COUNT);
1724 if (error) {
1725 /* Don't call itruncate_cleanup */
1726 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1727 xfs_trans_cancel(tp, 0);
1728 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1729 return VN_INACTIVE_CACHE;
1732 xfs_ilock(ip, XFS_ILOCK_EXCL);
1733 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1734 xfs_trans_ihold(tp, ip);
1737 * normally, we have to run xfs_itruncate_finish sync.
1738 * But if filesystem is wsync and we're in the inactive
1739 * path, then we know that nlink == 0, and that the
1740 * xaction that made nlink == 0 is permanently committed
1741 * since xfs_remove runs as a synchronous transaction.
1743 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1744 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1746 if (error) {
1747 xfs_trans_cancel(tp,
1748 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1749 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1750 return VN_INACTIVE_CACHE;
1752 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1755 * If we get an error while cleaning up a
1756 * symlink we bail out.
1758 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1759 xfs_inactive_symlink_rmt(ip, &tp) :
1760 xfs_inactive_symlink_local(ip, &tp);
1762 if (error) {
1763 ASSERT(tp == NULL);
1764 return VN_INACTIVE_CACHE;
1767 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1768 xfs_trans_ihold(tp, ip);
1769 } else {
1770 error = xfs_trans_reserve(tp, 0,
1771 XFS_IFREE_LOG_RES(mp),
1772 0, XFS_TRANS_PERM_LOG_RES,
1773 XFS_INACTIVE_LOG_COUNT);
1774 if (error) {
1775 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1776 xfs_trans_cancel(tp, 0);
1777 return VN_INACTIVE_CACHE;
1780 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1781 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1782 xfs_trans_ihold(tp, ip);
1786 * If there are attributes associated with the file
1787 * then blow them away now. The code calls a routine
1788 * that recursively deconstructs the attribute fork.
1789 * We need to just commit the current transaction
1790 * because we can't use it for xfs_attr_inactive().
1792 if (ip->i_d.di_anextents > 0) {
1793 error = xfs_inactive_attrs(ip, &tp);
1795 * If we got an error, the transaction is already
1796 * cancelled, and the inode is unlocked. Just get out.
1798 if (error)
1799 return VN_INACTIVE_CACHE;
1800 } else if (ip->i_afp) {
1801 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1805 * Free the inode.
1807 XFS_BMAP_INIT(&free_list, &first_block);
1808 error = xfs_ifree(tp, ip, &free_list);
1809 if (error) {
1811 * If we fail to free the inode, shut down. The cancel
1812 * might do that, we need to make sure. Otherwise the
1813 * inode might be lost for a long time or forever.
1815 if (!XFS_FORCED_SHUTDOWN(mp)) {
1816 cmn_err(CE_NOTE,
1817 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1818 error, mp->m_fsname);
1819 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1821 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1822 } else {
1824 * Credit the quota account(s). The inode is gone.
1826 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1829 * Just ignore errors at this point. There is
1830 * nothing we can do except to try to keep going.
1832 (void) xfs_bmap_finish(&tp, &free_list, &committed);
1833 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1836 * Release the dquots held by inode, if any.
1838 XFS_QM_DQDETACH(mp, ip);
1840 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1842 out:
1843 return VN_INACTIVE_CACHE;
1848 * xfs_lookup
1850 STATIC int
1851 xfs_lookup(
1852 bhv_desc_t *dir_bdp,
1853 bhv_vname_t *dentry,
1854 bhv_vnode_t **vpp,
1855 int flags,
1856 bhv_vnode_t *rdir,
1857 cred_t *credp)
1859 xfs_inode_t *dp, *ip;
1860 xfs_ino_t e_inum;
1861 int error;
1862 uint lock_mode;
1863 bhv_vnode_t *dir_vp;
1865 dir_vp = BHV_TO_VNODE(dir_bdp);
1866 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1868 dp = XFS_BHVTOI(dir_bdp);
1870 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1871 return XFS_ERROR(EIO);
1873 lock_mode = xfs_ilock_map_shared(dp);
1874 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1875 if (!error) {
1876 *vpp = XFS_ITOV(ip);
1877 ITRACE(ip);
1879 xfs_iunlock_map_shared(dp, lock_mode);
1880 return error;
1885 * xfs_create (create a new file).
1887 STATIC int
1888 xfs_create(
1889 bhv_desc_t *dir_bdp,
1890 bhv_vname_t *dentry,
1891 bhv_vattr_t *vap,
1892 bhv_vnode_t **vpp,
1893 cred_t *credp)
1895 char *name = VNAME(dentry);
1896 bhv_vnode_t *dir_vp;
1897 xfs_inode_t *dp, *ip;
1898 bhv_vnode_t *vp = NULL;
1899 xfs_trans_t *tp;
1900 xfs_mount_t *mp;
1901 xfs_dev_t rdev;
1902 int error;
1903 xfs_bmap_free_t free_list;
1904 xfs_fsblock_t first_block;
1905 boolean_t dp_joined_to_trans;
1906 int dm_event_sent = 0;
1907 uint cancel_flags;
1908 int committed;
1909 xfs_prid_t prid;
1910 struct xfs_dquot *udqp, *gdqp;
1911 uint resblks;
1912 int dm_di_mode;
1913 int namelen;
1915 ASSERT(!*vpp);
1916 dir_vp = BHV_TO_VNODE(dir_bdp);
1917 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1919 dp = XFS_BHVTOI(dir_bdp);
1920 mp = dp->i_mount;
1922 dm_di_mode = vap->va_mode;
1923 namelen = VNAMELEN(dentry);
1925 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1926 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1927 dir_vp, DM_RIGHT_NULL, NULL,
1928 DM_RIGHT_NULL, name, NULL,
1929 dm_di_mode, 0, 0);
1931 if (error)
1932 return error;
1933 dm_event_sent = 1;
1936 if (XFS_FORCED_SHUTDOWN(mp))
1937 return XFS_ERROR(EIO);
1939 /* Return through std_return after this point. */
1941 udqp = gdqp = NULL;
1942 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1943 prid = dp->i_d.di_projid;
1944 else if (vap->va_mask & XFS_AT_PROJID)
1945 prid = (xfs_prid_t)vap->va_projid;
1946 else
1947 prid = (xfs_prid_t)dfltprid;
1950 * Make sure that we have allocated dquot(s) on disk.
1952 error = XFS_QM_DQVOPALLOC(mp, dp,
1953 current_fsuid(credp), current_fsgid(credp), prid,
1954 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1955 if (error)
1956 goto std_return;
1958 ip = NULL;
1959 dp_joined_to_trans = B_FALSE;
1961 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1962 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1963 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1965 * Initially assume that the file does not exist and
1966 * reserve the resources for that case. If that is not
1967 * the case we'll drop the one we have and get a more
1968 * appropriate transaction later.
1970 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1971 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1972 if (error == ENOSPC) {
1973 resblks = 0;
1974 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1975 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1977 if (error) {
1978 cancel_flags = 0;
1979 dp = NULL;
1980 goto error_return;
1983 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1985 XFS_BMAP_INIT(&free_list, &first_block);
1987 ASSERT(ip == NULL);
1990 * Reserve disk quota and the inode.
1992 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1993 if (error)
1994 goto error_return;
1996 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
1997 goto error_return;
1998 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1999 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
2000 rdev, credp, prid, resblks > 0,
2001 &ip, &committed);
2002 if (error) {
2003 if (error == ENOSPC)
2004 goto error_return;
2005 goto abort_return;
2007 ITRACE(ip);
2010 * At this point, we've gotten a newly allocated inode.
2011 * It is locked (and joined to the transaction).
2014 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
2017 * Now we join the directory inode to the transaction.
2018 * We do not do it earlier because xfs_dir_ialloc
2019 * might commit the previous transaction (and release
2020 * all the locks).
2023 VN_HOLD(dir_vp);
2024 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2025 dp_joined_to_trans = B_TRUE;
2027 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
2028 &first_block, &free_list, resblks ?
2029 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2030 if (error) {
2031 ASSERT(error != ENOSPC);
2032 goto abort_return;
2034 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2035 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2038 * If this is a synchronous mount, make sure that the
2039 * create transaction goes to disk before returning to
2040 * the user.
2042 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2043 xfs_trans_set_sync(tp);
2046 dp->i_gen++;
2049 * Attach the dquot(s) to the inodes and modify them incore.
2050 * These ids of the inode couldn't have changed since the new
2051 * inode has been locked ever since it was created.
2053 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2056 * xfs_trans_commit normally decrements the vnode ref count
2057 * when it unlocks the inode. Since we want to return the
2058 * vnode to the caller, we bump the vnode ref count now.
2060 IHOLD(ip);
2061 vp = XFS_ITOV(ip);
2063 error = xfs_bmap_finish(&tp, &free_list, &committed);
2064 if (error) {
2065 xfs_bmap_cancel(&free_list);
2066 goto abort_rele;
2069 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2070 if (error) {
2071 IRELE(ip);
2072 tp = NULL;
2073 goto error_return;
2076 XFS_QM_DQRELE(mp, udqp);
2077 XFS_QM_DQRELE(mp, gdqp);
2080 * Propagate the fact that the vnode changed after the
2081 * xfs_inode locks have been released.
2083 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2085 *vpp = vp;
2087 /* Fallthrough to std_return with error = 0 */
2089 std_return:
2090 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2091 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2092 DM_EVENT_POSTCREATE)) {
2093 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2094 dir_vp, DM_RIGHT_NULL,
2095 *vpp ? vp:NULL,
2096 DM_RIGHT_NULL, name, NULL,
2097 dm_di_mode, error, 0);
2099 return error;
2101 abort_return:
2102 cancel_flags |= XFS_TRANS_ABORT;
2103 /* FALLTHROUGH */
2105 error_return:
2106 if (tp != NULL)
2107 xfs_trans_cancel(tp, cancel_flags);
2109 if (!dp_joined_to_trans && (dp != NULL))
2110 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2111 XFS_QM_DQRELE(mp, udqp);
2112 XFS_QM_DQRELE(mp, gdqp);
2114 goto std_return;
2116 abort_rele:
2118 * Wait until after the current transaction is aborted to
2119 * release the inode. This prevents recursive transactions
2120 * and deadlocks from xfs_inactive.
2122 cancel_flags |= XFS_TRANS_ABORT;
2123 xfs_trans_cancel(tp, cancel_flags);
2124 IRELE(ip);
2126 XFS_QM_DQRELE(mp, udqp);
2127 XFS_QM_DQRELE(mp, gdqp);
2129 goto std_return;
2132 #ifdef DEBUG
2134 * Some counters to see if (and how often) we are hitting some deadlock
2135 * prevention code paths.
2138 int xfs_rm_locks;
2139 int xfs_rm_lock_delays;
2140 int xfs_rm_attempts;
2141 #endif
2144 * The following routine will lock the inodes associated with the
2145 * directory and the named entry in the directory. The locks are
2146 * acquired in increasing inode number.
2148 * If the entry is "..", then only the directory is locked. The
2149 * vnode ref count will still include that from the .. entry in
2150 * this case.
2152 * There is a deadlock we need to worry about. If the locked directory is
2153 * in the AIL, it might be blocking up the log. The next inode we lock
2154 * could be already locked by another thread waiting for log space (e.g
2155 * a permanent log reservation with a long running transaction (see
2156 * xfs_itruncate_finish)). To solve this, we must check if the directory
2157 * is in the ail and use lock_nowait. If we can't lock, we need to
2158 * drop the inode lock on the directory and try again. xfs_iunlock will
2159 * potentially push the tail if we were holding up the log.
2161 STATIC int
2162 xfs_lock_dir_and_entry(
2163 xfs_inode_t *dp,
2164 xfs_inode_t *ip) /* inode of entry 'name' */
2166 int attempts;
2167 xfs_ino_t e_inum;
2168 xfs_inode_t *ips[2];
2169 xfs_log_item_t *lp;
2171 #ifdef DEBUG
2172 xfs_rm_locks++;
2173 #endif
2174 attempts = 0;
2176 again:
2177 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2179 e_inum = ip->i_ino;
2181 ITRACE(ip);
2184 * We want to lock in increasing inum. Since we've already
2185 * acquired the lock on the directory, we may need to release
2186 * if if the inum of the entry turns out to be less.
2188 if (e_inum > dp->i_ino) {
2190 * We are already in the right order, so just
2191 * lock on the inode of the entry.
2192 * We need to use nowait if dp is in the AIL.
2195 lp = (xfs_log_item_t *)dp->i_itemp;
2196 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2197 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2198 attempts++;
2199 #ifdef DEBUG
2200 xfs_rm_attempts++;
2201 #endif
2204 * Unlock dp and try again.
2205 * xfs_iunlock will try to push the tail
2206 * if the inode is in the AIL.
2209 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2211 if ((attempts % 5) == 0) {
2212 delay(1); /* Don't just spin the CPU */
2213 #ifdef DEBUG
2214 xfs_rm_lock_delays++;
2215 #endif
2217 goto again;
2219 } else {
2220 xfs_ilock(ip, XFS_ILOCK_EXCL);
2222 } else if (e_inum < dp->i_ino) {
2223 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2225 ips[0] = ip;
2226 ips[1] = dp;
2227 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2229 /* else e_inum == dp->i_ino */
2230 /* This can happen if we're asked to lock /x/..
2231 * the entry is "..", which is also the parent directory.
2234 return 0;
2237 #ifdef DEBUG
2238 int xfs_locked_n;
2239 int xfs_small_retries;
2240 int xfs_middle_retries;
2241 int xfs_lots_retries;
2242 int xfs_lock_delays;
2243 #endif
2246 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2247 * a different value
2249 static inline int
2250 xfs_lock_inumorder(int lock_mode, int subclass)
2252 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
2253 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
2254 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
2255 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
2257 return lock_mode;
2261 * The following routine will lock n inodes in exclusive mode.
2262 * We assume the caller calls us with the inodes in i_ino order.
2264 * We need to detect deadlock where an inode that we lock
2265 * is in the AIL and we start waiting for another inode that is locked
2266 * by a thread in a long running transaction (such as truncate). This can
2267 * result in deadlock since the long running trans might need to wait
2268 * for the inode we just locked in order to push the tail and free space
2269 * in the log.
2271 void
2272 xfs_lock_inodes(
2273 xfs_inode_t **ips,
2274 int inodes,
2275 int first_locked,
2276 uint lock_mode)
2278 int attempts = 0, i, j, try_lock;
2279 xfs_log_item_t *lp;
2281 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2283 if (first_locked) {
2284 try_lock = 1;
2285 i = 1;
2286 } else {
2287 try_lock = 0;
2288 i = 0;
2291 again:
2292 for (; i < inodes; i++) {
2293 ASSERT(ips[i]);
2295 if (i && (ips[i] == ips[i-1])) /* Already locked */
2296 continue;
2299 * If try_lock is not set yet, make sure all locked inodes
2300 * are not in the AIL.
2301 * If any are, set try_lock to be used later.
2304 if (!try_lock) {
2305 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2306 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2307 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2308 try_lock++;
2314 * If any of the previous locks we have locked is in the AIL,
2315 * we must TRY to get the second and subsequent locks. If
2316 * we can't get any, we must release all we have
2317 * and try again.
2320 if (try_lock) {
2321 /* try_lock must be 0 if i is 0. */
2323 * try_lock means we have an inode locked
2324 * that is in the AIL.
2326 ASSERT(i != 0);
2327 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
2328 attempts++;
2331 * Unlock all previous guys and try again.
2332 * xfs_iunlock will try to push the tail
2333 * if the inode is in the AIL.
2336 for(j = i - 1; j >= 0; j--) {
2339 * Check to see if we've already
2340 * unlocked this one.
2341 * Not the first one going back,
2342 * and the inode ptr is the same.
2344 if ((j != (i - 1)) && ips[j] ==
2345 ips[j+1])
2346 continue;
2348 xfs_iunlock(ips[j], lock_mode);
2351 if ((attempts % 5) == 0) {
2352 delay(1); /* Don't just spin the CPU */
2353 #ifdef DEBUG
2354 xfs_lock_delays++;
2355 #endif
2357 i = 0;
2358 try_lock = 0;
2359 goto again;
2361 } else {
2362 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
2366 #ifdef DEBUG
2367 if (attempts) {
2368 if (attempts < 5) xfs_small_retries++;
2369 else if (attempts < 100) xfs_middle_retries++;
2370 else xfs_lots_retries++;
2371 } else {
2372 xfs_locked_n++;
2374 #endif
2377 #ifdef DEBUG
2378 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2379 int remove_which_error_return = 0;
2380 #else /* ! DEBUG */
2381 #define REMOVE_DEBUG_TRACE(x)
2382 #endif /* ! DEBUG */
2386 * xfs_remove
2389 STATIC int
2390 xfs_remove(
2391 bhv_desc_t *dir_bdp,
2392 bhv_vname_t *dentry,
2393 cred_t *credp)
2395 bhv_vnode_t *dir_vp;
2396 char *name = VNAME(dentry);
2397 xfs_inode_t *dp, *ip;
2398 xfs_trans_t *tp = NULL;
2399 xfs_mount_t *mp;
2400 int error = 0;
2401 xfs_bmap_free_t free_list;
2402 xfs_fsblock_t first_block;
2403 int cancel_flags;
2404 int committed;
2405 int dm_di_mode = 0;
2406 int link_zero;
2407 uint resblks;
2408 int namelen;
2410 dir_vp = BHV_TO_VNODE(dir_bdp);
2411 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2413 dp = XFS_BHVTOI(dir_bdp);
2414 mp = dp->i_mount;
2416 if (XFS_FORCED_SHUTDOWN(mp))
2417 return XFS_ERROR(EIO);
2419 namelen = VNAMELEN(dentry);
2421 if (!xfs_get_dir_entry(dentry, &ip)) {
2422 dm_di_mode = ip->i_d.di_mode;
2423 IRELE(ip);
2426 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2427 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2428 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2429 name, NULL, dm_di_mode, 0, 0);
2430 if (error)
2431 return error;
2434 /* From this point on, return through std_return */
2435 ip = NULL;
2438 * We need to get a reference to ip before we get our log
2439 * reservation. The reason for this is that we cannot call
2440 * xfs_iget for an inode for which we do not have a reference
2441 * once we've acquired a log reservation. This is because the
2442 * inode we are trying to get might be in xfs_inactive going
2443 * for a log reservation. Since we'll have to wait for the
2444 * inactive code to complete before returning from xfs_iget,
2445 * we need to make sure that we don't have log space reserved
2446 * when we call xfs_iget. Instead we get an unlocked reference
2447 * to the inode before getting our log reservation.
2449 error = xfs_get_dir_entry(dentry, &ip);
2450 if (error) {
2451 REMOVE_DEBUG_TRACE(__LINE__);
2452 goto std_return;
2455 dm_di_mode = ip->i_d.di_mode;
2457 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2459 ITRACE(ip);
2461 error = XFS_QM_DQATTACH(mp, dp, 0);
2462 if (!error && dp != ip)
2463 error = XFS_QM_DQATTACH(mp, ip, 0);
2464 if (error) {
2465 REMOVE_DEBUG_TRACE(__LINE__);
2466 IRELE(ip);
2467 goto std_return;
2470 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2471 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2473 * We try to get the real space reservation first,
2474 * allowing for directory btree deletion(s) implying
2475 * possible bmap insert(s). If we can't get the space
2476 * reservation then we use 0 instead, and avoid the bmap
2477 * btree insert(s) in the directory code by, if the bmap
2478 * insert tries to happen, instead trimming the LAST
2479 * block from the directory.
2481 resblks = XFS_REMOVE_SPACE_RES(mp);
2482 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2483 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2484 if (error == ENOSPC) {
2485 resblks = 0;
2486 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2487 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2489 if (error) {
2490 ASSERT(error != ENOSPC);
2491 REMOVE_DEBUG_TRACE(__LINE__);
2492 xfs_trans_cancel(tp, 0);
2493 IRELE(ip);
2494 return error;
2497 error = xfs_lock_dir_and_entry(dp, ip);
2498 if (error) {
2499 REMOVE_DEBUG_TRACE(__LINE__);
2500 xfs_trans_cancel(tp, cancel_flags);
2501 IRELE(ip);
2502 goto std_return;
2506 * At this point, we've gotten both the directory and the entry
2507 * inodes locked.
2509 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2510 if (dp != ip) {
2512 * Increment vnode ref count only in this case since
2513 * there's an extra vnode reference in the case where
2514 * dp == ip.
2516 IHOLD(dp);
2517 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2521 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2523 XFS_BMAP_INIT(&free_list, &first_block);
2524 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2525 &first_block, &free_list, 0);
2526 if (error) {
2527 ASSERT(error != ENOENT);
2528 REMOVE_DEBUG_TRACE(__LINE__);
2529 goto error1;
2531 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2533 dp->i_gen++;
2534 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2536 error = xfs_droplink(tp, ip);
2537 if (error) {
2538 REMOVE_DEBUG_TRACE(__LINE__);
2539 goto error1;
2542 /* Determine if this is the last link while
2543 * we are in the transaction.
2545 link_zero = (ip)->i_d.di_nlink==0;
2548 * Take an extra ref on the inode so that it doesn't
2549 * go to xfs_inactive() from within the commit.
2551 IHOLD(ip);
2554 * If this is a synchronous mount, make sure that the
2555 * remove transaction goes to disk before returning to
2556 * the user.
2558 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2559 xfs_trans_set_sync(tp);
2562 error = xfs_bmap_finish(&tp, &free_list, &committed);
2563 if (error) {
2564 REMOVE_DEBUG_TRACE(__LINE__);
2565 goto error_rele;
2568 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2569 if (error) {
2570 IRELE(ip);
2571 goto std_return;
2575 * Before we drop our extra reference to the inode, purge it
2576 * from the refcache if it is there. By waiting until afterwards
2577 * to do the IRELE, we ensure that we won't go inactive in the
2578 * xfs_refcache_purge_ip routine (although that would be OK).
2580 xfs_refcache_purge_ip(ip);
2583 * If we are using filestreams, kill the stream association.
2584 * If the file is still open it may get a new one but that
2585 * will get killed on last close in xfs_close() so we don't
2586 * have to worry about that.
2588 if (link_zero && xfs_inode_is_filestream(ip))
2589 xfs_filestream_deassociate(ip);
2591 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2594 * Let interposed file systems know about removed links.
2596 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
2598 IRELE(ip);
2600 /* Fall through to std_return with error = 0 */
2601 std_return:
2602 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2603 DM_EVENT_POSTREMOVE)) {
2604 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2605 dir_vp, DM_RIGHT_NULL,
2606 NULL, DM_RIGHT_NULL,
2607 name, NULL, dm_di_mode, error, 0);
2609 return error;
2611 error1:
2612 xfs_bmap_cancel(&free_list);
2613 cancel_flags |= XFS_TRANS_ABORT;
2614 xfs_trans_cancel(tp, cancel_flags);
2615 goto std_return;
2617 error_rele:
2619 * In this case make sure to not release the inode until after
2620 * the current transaction is aborted. Releasing it beforehand
2621 * can cause us to go to xfs_inactive and start a recursive
2622 * transaction which can easily deadlock with the current one.
2624 xfs_bmap_cancel(&free_list);
2625 cancel_flags |= XFS_TRANS_ABORT;
2626 xfs_trans_cancel(tp, cancel_flags);
2629 * Before we drop our extra reference to the inode, purge it
2630 * from the refcache if it is there. By waiting until afterwards
2631 * to do the IRELE, we ensure that we won't go inactive in the
2632 * xfs_refcache_purge_ip routine (although that would be OK).
2634 xfs_refcache_purge_ip(ip);
2636 IRELE(ip);
2638 goto std_return;
2643 * xfs_link
2646 STATIC int
2647 xfs_link(
2648 bhv_desc_t *target_dir_bdp,
2649 bhv_vnode_t *src_vp,
2650 bhv_vname_t *dentry,
2651 cred_t *credp)
2653 xfs_inode_t *tdp, *sip;
2654 xfs_trans_t *tp;
2655 xfs_mount_t *mp;
2656 xfs_inode_t *ips[2];
2657 int error;
2658 xfs_bmap_free_t free_list;
2659 xfs_fsblock_t first_block;
2660 int cancel_flags;
2661 int committed;
2662 bhv_vnode_t *target_dir_vp;
2663 int resblks;
2664 char *target_name = VNAME(dentry);
2665 int target_namelen;
2667 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2668 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2669 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2671 target_namelen = VNAMELEN(dentry);
2672 ASSERT(!VN_ISDIR(src_vp));
2674 sip = xfs_vtoi(src_vp);
2675 tdp = XFS_BHVTOI(target_dir_bdp);
2676 mp = tdp->i_mount;
2677 if (XFS_FORCED_SHUTDOWN(mp))
2678 return XFS_ERROR(EIO);
2680 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2681 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2682 target_dir_vp, DM_RIGHT_NULL,
2683 src_vp, DM_RIGHT_NULL,
2684 target_name, NULL, 0, 0, 0);
2685 if (error)
2686 return error;
2689 /* Return through std_return after this point. */
2691 error = XFS_QM_DQATTACH(mp, sip, 0);
2692 if (!error && sip != tdp)
2693 error = XFS_QM_DQATTACH(mp, tdp, 0);
2694 if (error)
2695 goto std_return;
2697 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2698 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2699 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2700 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2701 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2702 if (error == ENOSPC) {
2703 resblks = 0;
2704 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2705 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2707 if (error) {
2708 cancel_flags = 0;
2709 goto error_return;
2712 if (sip->i_ino < tdp->i_ino) {
2713 ips[0] = sip;
2714 ips[1] = tdp;
2715 } else {
2716 ips[0] = tdp;
2717 ips[1] = sip;
2720 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2723 * Increment vnode ref counts since xfs_trans_commit &
2724 * xfs_trans_cancel will both unlock the inodes and
2725 * decrement the associated ref counts.
2727 VN_HOLD(src_vp);
2728 VN_HOLD(target_dir_vp);
2729 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2730 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2733 * If the source has too many links, we can't make any more to it.
2735 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2736 error = XFS_ERROR(EMLINK);
2737 goto error_return;
2741 * If we are using project inheritance, we only allow hard link
2742 * creation in our tree when the project IDs are the same; else
2743 * the tree quota mechanism could be circumvented.
2745 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2746 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2747 error = XFS_ERROR(EXDEV);
2748 goto error_return;
2751 if (resblks == 0 &&
2752 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
2753 goto error_return;
2755 XFS_BMAP_INIT(&free_list, &first_block);
2757 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
2758 sip->i_ino, &first_block, &free_list,
2759 resblks);
2760 if (error)
2761 goto abort_return;
2762 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2763 tdp->i_gen++;
2764 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2766 error = xfs_bumplink(tp, sip);
2767 if (error)
2768 goto abort_return;
2771 * If this is a synchronous mount, make sure that the
2772 * link transaction goes to disk before returning to
2773 * the user.
2775 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2776 xfs_trans_set_sync(tp);
2779 error = xfs_bmap_finish (&tp, &free_list, &committed);
2780 if (error) {
2781 xfs_bmap_cancel(&free_list);
2782 goto abort_return;
2785 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2786 if (error)
2787 goto std_return;
2789 /* Fall through to std_return with error = 0. */
2790 std_return:
2791 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2792 DM_EVENT_POSTLINK)) {
2793 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2794 target_dir_vp, DM_RIGHT_NULL,
2795 src_vp, DM_RIGHT_NULL,
2796 target_name, NULL, 0, error, 0);
2798 return error;
2800 abort_return:
2801 cancel_flags |= XFS_TRANS_ABORT;
2802 /* FALLTHROUGH */
2804 error_return:
2805 xfs_trans_cancel(tp, cancel_flags);
2806 goto std_return;
2811 * xfs_mkdir
2814 STATIC int
2815 xfs_mkdir(
2816 bhv_desc_t *dir_bdp,
2817 bhv_vname_t *dentry,
2818 bhv_vattr_t *vap,
2819 bhv_vnode_t **vpp,
2820 cred_t *credp)
2822 char *dir_name = VNAME(dentry);
2823 xfs_inode_t *dp;
2824 xfs_inode_t *cdp; /* inode of created dir */
2825 bhv_vnode_t *cvp; /* vnode of created dir */
2826 xfs_trans_t *tp;
2827 xfs_mount_t *mp;
2828 int cancel_flags;
2829 int error;
2830 int committed;
2831 xfs_bmap_free_t free_list;
2832 xfs_fsblock_t first_block;
2833 bhv_vnode_t *dir_vp;
2834 boolean_t dp_joined_to_trans;
2835 boolean_t created = B_FALSE;
2836 int dm_event_sent = 0;
2837 xfs_prid_t prid;
2838 struct xfs_dquot *udqp, *gdqp;
2839 uint resblks;
2840 int dm_di_mode;
2841 int dir_namelen;
2843 dir_vp = BHV_TO_VNODE(dir_bdp);
2844 dp = XFS_BHVTOI(dir_bdp);
2845 mp = dp->i_mount;
2847 if (XFS_FORCED_SHUTDOWN(mp))
2848 return XFS_ERROR(EIO);
2850 dir_namelen = VNAMELEN(dentry);
2852 tp = NULL;
2853 dp_joined_to_trans = B_FALSE;
2854 dm_di_mode = vap->va_mode;
2856 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2857 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2858 dir_vp, DM_RIGHT_NULL, NULL,
2859 DM_RIGHT_NULL, dir_name, NULL,
2860 dm_di_mode, 0, 0);
2861 if (error)
2862 return error;
2863 dm_event_sent = 1;
2866 /* Return through std_return after this point. */
2868 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2870 mp = dp->i_mount;
2871 udqp = gdqp = NULL;
2872 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2873 prid = dp->i_d.di_projid;
2874 else if (vap->va_mask & XFS_AT_PROJID)
2875 prid = (xfs_prid_t)vap->va_projid;
2876 else
2877 prid = (xfs_prid_t)dfltprid;
2880 * Make sure that we have allocated dquot(s) on disk.
2882 error = XFS_QM_DQVOPALLOC(mp, dp,
2883 current_fsuid(credp), current_fsgid(credp), prid,
2884 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2885 if (error)
2886 goto std_return;
2888 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2889 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2890 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2891 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2892 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2893 if (error == ENOSPC) {
2894 resblks = 0;
2895 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2896 XFS_TRANS_PERM_LOG_RES,
2897 XFS_MKDIR_LOG_COUNT);
2899 if (error) {
2900 cancel_flags = 0;
2901 dp = NULL;
2902 goto error_return;
2905 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2908 * Check for directory link count overflow.
2910 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2911 error = XFS_ERROR(EMLINK);
2912 goto error_return;
2916 * Reserve disk quota and the inode.
2918 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2919 if (error)
2920 goto error_return;
2922 if (resblks == 0 &&
2923 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
2924 goto error_return;
2926 * create the directory inode.
2928 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2929 0, credp, prid, resblks > 0,
2930 &cdp, NULL);
2931 if (error) {
2932 if (error == ENOSPC)
2933 goto error_return;
2934 goto abort_return;
2936 ITRACE(cdp);
2939 * Now we add the directory inode to the transaction.
2940 * We waited until now since xfs_dir_ialloc might start
2941 * a new transaction. Had we joined the transaction
2942 * earlier, the locks might have gotten released.
2944 VN_HOLD(dir_vp);
2945 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2946 dp_joined_to_trans = B_TRUE;
2948 XFS_BMAP_INIT(&free_list, &first_block);
2950 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2951 &first_block, &free_list, resblks ?
2952 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2953 if (error) {
2954 ASSERT(error != ENOSPC);
2955 goto error1;
2957 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2960 * Bump the in memory version number of the parent directory
2961 * so that other processes accessing it will recognize that
2962 * the directory has changed.
2964 dp->i_gen++;
2966 error = xfs_dir_init(tp, cdp, dp);
2967 if (error)
2968 goto error2;
2970 cdp->i_gen = 1;
2971 error = xfs_bumplink(tp, dp);
2972 if (error)
2973 goto error2;
2975 cvp = XFS_ITOV(cdp);
2977 created = B_TRUE;
2979 *vpp = cvp;
2980 IHOLD(cdp);
2983 * Attach the dquots to the new inode and modify the icount incore.
2985 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2988 * If this is a synchronous mount, make sure that the
2989 * mkdir transaction goes to disk before returning to
2990 * the user.
2992 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2993 xfs_trans_set_sync(tp);
2996 error = xfs_bmap_finish(&tp, &free_list, &committed);
2997 if (error) {
2998 IRELE(cdp);
2999 goto error2;
3002 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3003 XFS_QM_DQRELE(mp, udqp);
3004 XFS_QM_DQRELE(mp, gdqp);
3005 if (error) {
3006 IRELE(cdp);
3009 /* Fall through to std_return with error = 0 or errno from
3010 * xfs_trans_commit. */
3012 std_return:
3013 if ( (created || (error != 0 && dm_event_sent != 0)) &&
3014 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3015 DM_EVENT_POSTCREATE)) {
3016 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
3017 dir_vp, DM_RIGHT_NULL,
3018 created ? XFS_ITOV(cdp):NULL,
3019 DM_RIGHT_NULL,
3020 dir_name, NULL,
3021 dm_di_mode, error, 0);
3023 return error;
3025 error2:
3026 error1:
3027 xfs_bmap_cancel(&free_list);
3028 abort_return:
3029 cancel_flags |= XFS_TRANS_ABORT;
3030 error_return:
3031 xfs_trans_cancel(tp, cancel_flags);
3032 XFS_QM_DQRELE(mp, udqp);
3033 XFS_QM_DQRELE(mp, gdqp);
3035 if (!dp_joined_to_trans && (dp != NULL)) {
3036 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3039 goto std_return;
3044 * xfs_rmdir
3047 STATIC int
3048 xfs_rmdir(
3049 bhv_desc_t *dir_bdp,
3050 bhv_vname_t *dentry,
3051 cred_t *credp)
3053 char *name = VNAME(dentry);
3054 xfs_inode_t *dp;
3055 xfs_inode_t *cdp; /* child directory */
3056 xfs_trans_t *tp;
3057 xfs_mount_t *mp;
3058 int error;
3059 xfs_bmap_free_t free_list;
3060 xfs_fsblock_t first_block;
3061 int cancel_flags;
3062 int committed;
3063 bhv_vnode_t *dir_vp;
3064 int dm_di_mode = S_IFDIR;
3065 int last_cdp_link;
3066 int namelen;
3067 uint resblks;
3069 dir_vp = BHV_TO_VNODE(dir_bdp);
3070 dp = XFS_BHVTOI(dir_bdp);
3071 mp = dp->i_mount;
3073 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3075 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3076 return XFS_ERROR(EIO);
3077 namelen = VNAMELEN(dentry);
3079 if (!xfs_get_dir_entry(dentry, &cdp)) {
3080 dm_di_mode = cdp->i_d.di_mode;
3081 IRELE(cdp);
3084 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3085 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3086 dir_vp, DM_RIGHT_NULL,
3087 NULL, DM_RIGHT_NULL,
3088 name, NULL, dm_di_mode, 0, 0);
3089 if (error)
3090 return XFS_ERROR(error);
3093 /* Return through std_return after this point. */
3095 cdp = NULL;
3098 * We need to get a reference to cdp before we get our log
3099 * reservation. The reason for this is that we cannot call
3100 * xfs_iget for an inode for which we do not have a reference
3101 * once we've acquired a log reservation. This is because the
3102 * inode we are trying to get might be in xfs_inactive going
3103 * for a log reservation. Since we'll have to wait for the
3104 * inactive code to complete before returning from xfs_iget,
3105 * we need to make sure that we don't have log space reserved
3106 * when we call xfs_iget. Instead we get an unlocked reference
3107 * to the inode before getting our log reservation.
3109 error = xfs_get_dir_entry(dentry, &cdp);
3110 if (error) {
3111 REMOVE_DEBUG_TRACE(__LINE__);
3112 goto std_return;
3114 mp = dp->i_mount;
3115 dm_di_mode = cdp->i_d.di_mode;
3118 * Get the dquots for the inodes.
3120 error = XFS_QM_DQATTACH(mp, dp, 0);
3121 if (!error && dp != cdp)
3122 error = XFS_QM_DQATTACH(mp, cdp, 0);
3123 if (error) {
3124 IRELE(cdp);
3125 REMOVE_DEBUG_TRACE(__LINE__);
3126 goto std_return;
3129 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3130 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3132 * We try to get the real space reservation first,
3133 * allowing for directory btree deletion(s) implying
3134 * possible bmap insert(s). If we can't get the space
3135 * reservation then we use 0 instead, and avoid the bmap
3136 * btree insert(s) in the directory code by, if the bmap
3137 * insert tries to happen, instead trimming the LAST
3138 * block from the directory.
3140 resblks = XFS_REMOVE_SPACE_RES(mp);
3141 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3142 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3143 if (error == ENOSPC) {
3144 resblks = 0;
3145 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3146 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3148 if (error) {
3149 ASSERT(error != ENOSPC);
3150 cancel_flags = 0;
3151 IRELE(cdp);
3152 goto error_return;
3154 XFS_BMAP_INIT(&free_list, &first_block);
3157 * Now lock the child directory inode and the parent directory
3158 * inode in the proper order. This will take care of validating
3159 * that the directory entry for the child directory inode has
3160 * not changed while we were obtaining a log reservation.
3162 error = xfs_lock_dir_and_entry(dp, cdp);
3163 if (error) {
3164 xfs_trans_cancel(tp, cancel_flags);
3165 IRELE(cdp);
3166 goto std_return;
3169 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3170 if (dp != cdp) {
3172 * Only increment the parent directory vnode count if
3173 * we didn't bump it in looking up cdp. The only time
3174 * we don't bump it is when we're looking up ".".
3176 VN_HOLD(dir_vp);
3179 ITRACE(cdp);
3180 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3182 ASSERT(cdp->i_d.di_nlink >= 2);
3183 if (cdp->i_d.di_nlink != 2) {
3184 error = XFS_ERROR(ENOTEMPTY);
3185 goto error_return;
3187 if (!xfs_dir_isempty(cdp)) {
3188 error = XFS_ERROR(ENOTEMPTY);
3189 goto error_return;
3192 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
3193 &first_block, &free_list, resblks);
3194 if (error)
3195 goto error1;
3197 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3200 * Bump the in memory generation count on the parent
3201 * directory so that other can know that it has changed.
3203 dp->i_gen++;
3206 * Drop the link from cdp's "..".
3208 error = xfs_droplink(tp, dp);
3209 if (error) {
3210 goto error1;
3214 * Drop the link from dp to cdp.
3216 error = xfs_droplink(tp, cdp);
3217 if (error) {
3218 goto error1;
3222 * Drop the "." link from cdp to self.
3224 error = xfs_droplink(tp, cdp);
3225 if (error) {
3226 goto error1;
3229 /* Determine these before committing transaction */
3230 last_cdp_link = (cdp)->i_d.di_nlink==0;
3233 * Take an extra ref on the child vnode so that it
3234 * does not go to xfs_inactive() from within the commit.
3236 IHOLD(cdp);
3239 * If this is a synchronous mount, make sure that the
3240 * rmdir transaction goes to disk before returning to
3241 * the user.
3243 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3244 xfs_trans_set_sync(tp);
3247 error = xfs_bmap_finish (&tp, &free_list, &committed);
3248 if (error) {
3249 xfs_bmap_cancel(&free_list);
3250 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3251 XFS_TRANS_ABORT));
3252 IRELE(cdp);
3253 goto std_return;
3256 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3257 if (error) {
3258 IRELE(cdp);
3259 goto std_return;
3264 * Let interposed file systems know about removed links.
3266 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3268 IRELE(cdp);
3270 /* Fall through to std_return with error = 0 or the errno
3271 * from xfs_trans_commit. */
3272 std_return:
3273 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3274 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3275 dir_vp, DM_RIGHT_NULL,
3276 NULL, DM_RIGHT_NULL,
3277 name, NULL, dm_di_mode,
3278 error, 0);
3280 return error;
3282 error1:
3283 xfs_bmap_cancel(&free_list);
3284 cancel_flags |= XFS_TRANS_ABORT;
3285 /* FALLTHROUGH */
3287 error_return:
3288 xfs_trans_cancel(tp, cancel_flags);
3289 goto std_return;
3294 * Read dp's entries starting at uiop->uio_offset and translate them into
3295 * bufsize bytes worth of struct dirents starting at bufbase.
3297 STATIC int
3298 xfs_readdir(
3299 bhv_desc_t *dir_bdp,
3300 uio_t *uiop,
3301 cred_t *credp,
3302 int *eofp)
3304 xfs_inode_t *dp;
3305 xfs_trans_t *tp = NULL;
3306 int error = 0;
3307 uint lock_mode;
3309 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3310 (inst_t *)__return_address);
3311 dp = XFS_BHVTOI(dir_bdp);
3313 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
3314 return XFS_ERROR(EIO);
3316 lock_mode = xfs_ilock_map_shared(dp);
3317 error = xfs_dir_getdents(tp, dp, uiop, eofp);
3318 xfs_iunlock_map_shared(dp, lock_mode);
3319 return error;
3323 STATIC int
3324 xfs_symlink(
3325 bhv_desc_t *dir_bdp,
3326 bhv_vname_t *dentry,
3327 bhv_vattr_t *vap,
3328 char *target_path,
3329 bhv_vnode_t **vpp,
3330 cred_t *credp)
3332 xfs_trans_t *tp;
3333 xfs_mount_t *mp;
3334 xfs_inode_t *dp;
3335 xfs_inode_t *ip;
3336 int error;
3337 int pathlen;
3338 xfs_bmap_free_t free_list;
3339 xfs_fsblock_t first_block;
3340 boolean_t dp_joined_to_trans;
3341 bhv_vnode_t *dir_vp;
3342 uint cancel_flags;
3343 int committed;
3344 xfs_fileoff_t first_fsb;
3345 xfs_filblks_t fs_blocks;
3346 int nmaps;
3347 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3348 xfs_daddr_t d;
3349 char *cur_chunk;
3350 int byte_cnt;
3351 int n;
3352 xfs_buf_t *bp;
3353 xfs_prid_t prid;
3354 struct xfs_dquot *udqp, *gdqp;
3355 uint resblks;
3356 char *link_name = VNAME(dentry);
3357 int link_namelen;
3359 *vpp = NULL;
3360 dir_vp = BHV_TO_VNODE(dir_bdp);
3361 dp = XFS_BHVTOI(dir_bdp);
3362 dp_joined_to_trans = B_FALSE;
3363 error = 0;
3364 ip = NULL;
3365 tp = NULL;
3367 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3369 mp = dp->i_mount;
3371 if (XFS_FORCED_SHUTDOWN(mp))
3372 return XFS_ERROR(EIO);
3374 link_namelen = VNAMELEN(dentry);
3377 * Check component lengths of the target path name.
3379 pathlen = strlen(target_path);
3380 if (pathlen >= MAXPATHLEN) /* total string too long */
3381 return XFS_ERROR(ENAMETOOLONG);
3382 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3383 int len, total;
3384 char *path;
3386 for (total = 0, path = target_path; total < pathlen;) {
3388 * Skip any slashes.
3390 while(*path == '/') {
3391 total++;
3392 path++;
3396 * Count up to the next slash or end of path.
3397 * Error out if the component is bigger than MAXNAMELEN.
3399 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3400 if (++len >= MAXNAMELEN) {
3401 error = ENAMETOOLONG;
3402 return error;
3408 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3409 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3410 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3411 link_name, target_path, 0, 0, 0);
3412 if (error)
3413 return error;
3416 /* Return through std_return after this point. */
3418 udqp = gdqp = NULL;
3419 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3420 prid = dp->i_d.di_projid;
3421 else if (vap->va_mask & XFS_AT_PROJID)
3422 prid = (xfs_prid_t)vap->va_projid;
3423 else
3424 prid = (xfs_prid_t)dfltprid;
3427 * Make sure that we have allocated dquot(s) on disk.
3429 error = XFS_QM_DQVOPALLOC(mp, dp,
3430 current_fsuid(credp), current_fsgid(credp), prid,
3431 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3432 if (error)
3433 goto std_return;
3435 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3436 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3438 * The symlink will fit into the inode data fork?
3439 * There can't be any attributes so we get the whole variable part.
3441 if (pathlen <= XFS_LITINO(mp))
3442 fs_blocks = 0;
3443 else
3444 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3445 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3446 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3447 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3448 if (error == ENOSPC && fs_blocks == 0) {
3449 resblks = 0;
3450 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3451 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3453 if (error) {
3454 cancel_flags = 0;
3455 dp = NULL;
3456 goto error_return;
3459 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
3462 * Check whether the directory allows new symlinks or not.
3464 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3465 error = XFS_ERROR(EPERM);
3466 goto error_return;
3470 * Reserve disk quota : blocks and inode.
3472 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3473 if (error)
3474 goto error_return;
3477 * Check for ability to enter directory entry, if no space reserved.
3479 if (resblks == 0 &&
3480 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
3481 goto error_return;
3483 * Initialize the bmap freelist prior to calling either
3484 * bmapi or the directory create code.
3486 XFS_BMAP_INIT(&free_list, &first_block);
3489 * Allocate an inode for the symlink.
3491 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3492 1, 0, credp, prid, resblks > 0, &ip, NULL);
3493 if (error) {
3494 if (error == ENOSPC)
3495 goto error_return;
3496 goto error1;
3498 ITRACE(ip);
3500 VN_HOLD(dir_vp);
3501 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3502 dp_joined_to_trans = B_TRUE;
3505 * Also attach the dquot(s) to it, if applicable.
3507 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3509 if (resblks)
3510 resblks -= XFS_IALLOC_SPACE_RES(mp);
3512 * If the symlink will fit into the inode, write it inline.
3514 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3515 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3516 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3517 ip->i_d.di_size = pathlen;
3520 * The inode was initially created in extent format.
3522 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3523 ip->i_df.if_flags |= XFS_IFINLINE;
3525 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3526 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3528 } else {
3529 first_fsb = 0;
3530 nmaps = SYMLINK_MAPS;
3532 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3533 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3534 &first_block, resblks, mval, &nmaps,
3535 &free_list, NULL);
3536 if (error) {
3537 goto error1;
3540 if (resblks)
3541 resblks -= fs_blocks;
3542 ip->i_d.di_size = pathlen;
3543 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3545 cur_chunk = target_path;
3546 for (n = 0; n < nmaps; n++) {
3547 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3548 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3549 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3550 BTOBB(byte_cnt), 0);
3551 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3552 if (pathlen < byte_cnt) {
3553 byte_cnt = pathlen;
3555 pathlen -= byte_cnt;
3557 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3558 cur_chunk += byte_cnt;
3560 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3565 * Create the directory entry for the symlink.
3567 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3568 &first_block, &free_list, resblks);
3569 if (error)
3570 goto error1;
3571 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3572 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3575 * Bump the in memory version number of the parent directory
3576 * so that other processes accessing it will recognize that
3577 * the directory has changed.
3579 dp->i_gen++;
3582 * If this is a synchronous mount, make sure that the
3583 * symlink transaction goes to disk before returning to
3584 * the user.
3586 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3587 xfs_trans_set_sync(tp);
3591 * xfs_trans_commit normally decrements the vnode ref count
3592 * when it unlocks the inode. Since we want to return the
3593 * vnode to the caller, we bump the vnode ref count now.
3595 IHOLD(ip);
3597 error = xfs_bmap_finish(&tp, &free_list, &committed);
3598 if (error) {
3599 goto error2;
3601 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3602 XFS_QM_DQRELE(mp, udqp);
3603 XFS_QM_DQRELE(mp, gdqp);
3605 /* Fall through to std_return with error = 0 or errno from
3606 * xfs_trans_commit */
3607 std_return:
3608 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3609 DM_EVENT_POSTSYMLINK)) {
3610 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3611 dir_vp, DM_RIGHT_NULL,
3612 error ? NULL : XFS_ITOV(ip),
3613 DM_RIGHT_NULL, link_name, target_path,
3614 0, error, 0);
3617 if (!error) {
3618 bhv_vnode_t *vp;
3620 ASSERT(ip);
3621 vp = XFS_ITOV(ip);
3622 *vpp = vp;
3624 return error;
3626 error2:
3627 IRELE(ip);
3628 error1:
3629 xfs_bmap_cancel(&free_list);
3630 cancel_flags |= XFS_TRANS_ABORT;
3631 error_return:
3632 xfs_trans_cancel(tp, cancel_flags);
3633 XFS_QM_DQRELE(mp, udqp);
3634 XFS_QM_DQRELE(mp, gdqp);
3636 if (!dp_joined_to_trans && (dp != NULL)) {
3637 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3640 goto std_return;
3645 * xfs_fid2
3647 * A fid routine that takes a pointer to a previously allocated
3648 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3650 STATIC int
3651 xfs_fid2(
3652 bhv_desc_t *bdp,
3653 fid_t *fidp)
3655 xfs_inode_t *ip;
3656 xfs_fid2_t *xfid;
3658 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3659 (inst_t *)__return_address);
3660 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3662 xfid = (xfs_fid2_t *)fidp;
3663 ip = XFS_BHVTOI(bdp);
3664 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3665 xfid->fid_pad = 0;
3667 * use memcpy because the inode is a long long and there's no
3668 * assurance that xfid->fid_ino is properly aligned.
3670 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3671 xfid->fid_gen = ip->i_d.di_gen;
3673 return 0;
3678 * xfs_rwlock
3681 xfs_rwlock(
3682 bhv_desc_t *bdp,
3683 bhv_vrwlock_t locktype)
3685 xfs_inode_t *ip;
3686 bhv_vnode_t *vp;
3688 vp = BHV_TO_VNODE(bdp);
3689 if (VN_ISDIR(vp))
3690 return 1;
3691 ip = XFS_BHVTOI(bdp);
3692 if (locktype == VRWLOCK_WRITE) {
3693 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3694 } else if (locktype == VRWLOCK_TRY_READ) {
3695 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3696 } else if (locktype == VRWLOCK_TRY_WRITE) {
3697 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3698 } else {
3699 ASSERT((locktype == VRWLOCK_READ) ||
3700 (locktype == VRWLOCK_WRITE_DIRECT));
3701 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3704 return 1;
3709 * xfs_rwunlock
3711 void
3712 xfs_rwunlock(
3713 bhv_desc_t *bdp,
3714 bhv_vrwlock_t locktype)
3716 xfs_inode_t *ip;
3717 bhv_vnode_t *vp;
3719 vp = BHV_TO_VNODE(bdp);
3720 if (VN_ISDIR(vp))
3721 return;
3722 ip = XFS_BHVTOI(bdp);
3723 if (locktype == VRWLOCK_WRITE) {
3725 * In the write case, we may have added a new entry to
3726 * the reference cache. This might store a pointer to
3727 * an inode to be released in this inode. If it is there,
3728 * clear the pointer and release the inode after unlocking
3729 * this one.
3731 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3732 } else {
3733 ASSERT((locktype == VRWLOCK_READ) ||
3734 (locktype == VRWLOCK_WRITE_DIRECT));
3735 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3737 return;
3740 STATIC int
3741 xfs_inode_flush(
3742 bhv_desc_t *bdp,
3743 int flags)
3745 xfs_inode_t *ip;
3746 xfs_mount_t *mp;
3747 xfs_inode_log_item_t *iip;
3748 int error = 0;
3750 ip = XFS_BHVTOI(bdp);
3751 mp = ip->i_mount;
3752 iip = ip->i_itemp;
3754 if (XFS_FORCED_SHUTDOWN(mp))
3755 return XFS_ERROR(EIO);
3758 * Bypass inodes which have already been cleaned by
3759 * the inode flush clustering code inside xfs_iflush
3761 if ((ip->i_update_core == 0) &&
3762 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3763 return 0;
3765 if (flags & FLUSH_LOG) {
3766 if (iip && iip->ili_last_lsn) {
3767 xlog_t *log = mp->m_log;
3768 xfs_lsn_t sync_lsn;
3769 int s, log_flags = XFS_LOG_FORCE;
3771 s = GRANT_LOCK(log);
3772 sync_lsn = log->l_last_sync_lsn;
3773 GRANT_UNLOCK(log, s);
3775 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) > 0)) {
3776 if (flags & FLUSH_SYNC)
3777 log_flags |= XFS_LOG_SYNC;
3778 error = xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3779 if (error)
3780 return error;
3783 if (ip->i_update_core == 0)
3784 return 0;
3789 * We make this non-blocking if the inode is contended,
3790 * return EAGAIN to indicate to the caller that they
3791 * did not succeed. This prevents the flush path from
3792 * blocking on inodes inside another operation right
3793 * now, they get caught later by xfs_sync.
3795 if (flags & FLUSH_INODE) {
3796 int flush_flags;
3798 if (flags & FLUSH_SYNC) {
3799 xfs_ilock(ip, XFS_ILOCK_SHARED);
3800 xfs_iflock(ip);
3801 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3802 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3803 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3804 return EAGAIN;
3806 } else {
3807 return EAGAIN;
3810 if (flags & FLUSH_SYNC)
3811 flush_flags = XFS_IFLUSH_SYNC;
3812 else
3813 flush_flags = XFS_IFLUSH_ASYNC;
3815 error = xfs_iflush(ip, flush_flags);
3816 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3819 return error;
3823 xfs_set_dmattrs (
3824 bhv_desc_t *bdp,
3825 u_int evmask,
3826 u_int16_t state,
3827 cred_t *credp)
3829 xfs_inode_t *ip;
3830 xfs_trans_t *tp;
3831 xfs_mount_t *mp;
3832 int error;
3834 if (!capable(CAP_SYS_ADMIN))
3835 return XFS_ERROR(EPERM);
3837 ip = XFS_BHVTOI(bdp);
3838 mp = ip->i_mount;
3840 if (XFS_FORCED_SHUTDOWN(mp))
3841 return XFS_ERROR(EIO);
3843 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3844 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3845 if (error) {
3846 xfs_trans_cancel(tp, 0);
3847 return error;
3849 xfs_ilock(ip, XFS_ILOCK_EXCL);
3850 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3852 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3853 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3855 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3856 IHOLD(ip);
3857 error = xfs_trans_commit(tp, 0);
3859 return error;
3862 STATIC int
3863 xfs_reclaim(
3864 bhv_desc_t *bdp)
3866 xfs_inode_t *ip;
3867 bhv_vnode_t *vp;
3869 vp = BHV_TO_VNODE(bdp);
3870 ip = XFS_BHVTOI(bdp);
3872 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3874 ASSERT(!VN_MAPPED(vp));
3876 /* bad inode, get out here ASAP */
3877 if (VN_BAD(vp)) {
3878 xfs_ireclaim(ip);
3879 return 0;
3882 vn_iowait(vp);
3884 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3887 * Make sure the atime in the XFS inode is correct before freeing the
3888 * Linux inode.
3890 xfs_synchronize_atime(ip);
3893 * If we have nothing to flush with this inode then complete the
3894 * teardown now, otherwise break the link between the xfs inode and the
3895 * linux inode and clean up the xfs inode later. This avoids flushing
3896 * the inode to disk during the delete operation itself.
3898 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3899 * first to ensure that xfs_iunpin() will never see an xfs inode
3900 * that has a linux inode being reclaimed. Synchronisation is provided
3901 * by the i_flags_lock.
3903 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3904 xfs_ilock(ip, XFS_ILOCK_EXCL);
3905 xfs_iflock(ip);
3906 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3907 } else {
3908 xfs_mount_t *mp = ip->i_mount;
3910 /* Protect sync and unpin from us */
3911 XFS_MOUNT_ILOCK(mp);
3912 spin_lock(&ip->i_flags_lock);
3913 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
3914 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3915 spin_unlock(&ip->i_flags_lock);
3916 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3917 XFS_MOUNT_IUNLOCK(mp);
3919 return 0;
3923 xfs_finish_reclaim(
3924 xfs_inode_t *ip,
3925 int locked,
3926 int sync_mode)
3928 xfs_ihash_t *ih = ip->i_hash;
3929 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3930 int error;
3932 if (vp && VN_BAD(vp))
3933 goto reclaim;
3935 /* The hash lock here protects a thread in xfs_iget_core from
3936 * racing with us on linking the inode back with a vnode.
3937 * Once we have the XFS_IRECLAIM flag set it will not touch
3938 * us.
3940 write_lock(&ih->ih_lock);
3941 spin_lock(&ip->i_flags_lock);
3942 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3943 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
3944 spin_unlock(&ip->i_flags_lock);
3945 write_unlock(&ih->ih_lock);
3946 if (locked) {
3947 xfs_ifunlock(ip);
3948 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3950 return 1;
3952 __xfs_iflags_set(ip, XFS_IRECLAIM);
3953 spin_unlock(&ip->i_flags_lock);
3954 write_unlock(&ih->ih_lock);
3957 * If the inode is still dirty, then flush it out. If the inode
3958 * is not in the AIL, then it will be OK to flush it delwri as
3959 * long as xfs_iflush() does not keep any references to the inode.
3960 * We leave that decision up to xfs_iflush() since it has the
3961 * knowledge of whether it's OK to simply do a delwri flush of
3962 * the inode or whether we need to wait until the inode is
3963 * pulled from the AIL.
3964 * We get the flush lock regardless, though, just to make sure
3965 * we don't free it while it is being flushed.
3967 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3968 if (!locked) {
3969 xfs_ilock(ip, XFS_ILOCK_EXCL);
3970 xfs_iflock(ip);
3973 if (ip->i_update_core ||
3974 ((ip->i_itemp != NULL) &&
3975 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3976 error = xfs_iflush(ip, sync_mode);
3978 * If we hit an error, typically because of filesystem
3979 * shutdown, we don't need to let vn_reclaim to know
3980 * because we're gonna reclaim the inode anyway.
3982 if (error) {
3983 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3984 goto reclaim;
3986 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3989 ASSERT(ip->i_update_core == 0);
3990 ASSERT(ip->i_itemp == NULL ||
3991 ip->i_itemp->ili_format.ilf_fields == 0);
3992 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3993 } else if (locked) {
3995 * We are not interested in doing an iflush if we're
3996 * in the process of shutting down the filesystem forcibly.
3997 * So, just reclaim the inode.
3999 xfs_ifunlock(ip);
4000 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4003 reclaim:
4004 xfs_ireclaim(ip);
4005 return 0;
4009 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
4011 int purged;
4012 xfs_inode_t *ip, *n;
4013 int done = 0;
4015 while (!done) {
4016 purged = 0;
4017 XFS_MOUNT_ILOCK(mp);
4018 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
4019 if (noblock) {
4020 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4021 continue;
4022 if (xfs_ipincount(ip) ||
4023 !xfs_iflock_nowait(ip)) {
4024 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4025 continue;
4028 XFS_MOUNT_IUNLOCK(mp);
4029 if (xfs_finish_reclaim(ip, noblock,
4030 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
4031 delay(1);
4032 purged = 1;
4033 break;
4036 done = !purged;
4039 XFS_MOUNT_IUNLOCK(mp);
4040 return 0;
4044 * xfs_alloc_file_space()
4045 * This routine allocates disk space for the given file.
4047 * If alloc_type == 0, this request is for an ALLOCSP type
4048 * request which will change the file size. In this case, no
4049 * DMAPI event will be generated by the call. A TRUNCATE event
4050 * will be generated later by xfs_setattr.
4052 * If alloc_type != 0, this request is for a RESVSP type
4053 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4054 * lower block boundary byte address is less than the file's
4055 * length.
4057 * RETURNS:
4058 * 0 on success
4059 * errno on error
4062 STATIC int
4063 xfs_alloc_file_space(
4064 xfs_inode_t *ip,
4065 xfs_off_t offset,
4066 xfs_off_t len,
4067 int alloc_type,
4068 int attr_flags)
4070 xfs_mount_t *mp = ip->i_mount;
4071 xfs_off_t count;
4072 xfs_filblks_t allocated_fsb;
4073 xfs_filblks_t allocatesize_fsb;
4074 xfs_extlen_t extsz, temp;
4075 xfs_fileoff_t startoffset_fsb;
4076 xfs_fsblock_t firstfsb;
4077 int nimaps;
4078 int bmapi_flag;
4079 int quota_flag;
4080 int rt;
4081 xfs_trans_t *tp;
4082 xfs_bmbt_irec_t imaps[1], *imapp;
4083 xfs_bmap_free_t free_list;
4084 uint qblocks, resblks, resrtextents;
4085 int committed;
4086 int error;
4088 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4090 if (XFS_FORCED_SHUTDOWN(mp))
4091 return XFS_ERROR(EIO);
4093 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4094 return error;
4096 if (len <= 0)
4097 return XFS_ERROR(EINVAL);
4099 rt = XFS_IS_REALTIME_INODE(ip);
4100 extsz = xfs_get_extsz_hint(ip);
4102 count = len;
4103 imapp = &imaps[0];
4104 nimaps = 1;
4105 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4106 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4107 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4109 /* Generate a DMAPI event if needed. */
4110 if (alloc_type != 0 && offset < ip->i_size &&
4111 (attr_flags&ATTR_DMI) == 0 &&
4112 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4113 xfs_off_t end_dmi_offset;
4115 end_dmi_offset = offset+len;
4116 if (end_dmi_offset > ip->i_size)
4117 end_dmi_offset = ip->i_size;
4118 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4119 offset, end_dmi_offset - offset,
4120 0, NULL);
4121 if (error)
4122 return error;
4126 * Allocate file space until done or until there is an error
4128 retry:
4129 while (allocatesize_fsb && !error) {
4130 xfs_fileoff_t s, e;
4133 * Determine space reservations for data/realtime.
4135 if (unlikely(extsz)) {
4136 s = startoffset_fsb;
4137 do_div(s, extsz);
4138 s *= extsz;
4139 e = startoffset_fsb + allocatesize_fsb;
4140 if ((temp = do_mod(startoffset_fsb, extsz)))
4141 e += temp;
4142 if ((temp = do_mod(e, extsz)))
4143 e += extsz - temp;
4144 } else {
4145 s = 0;
4146 e = allocatesize_fsb;
4149 if (unlikely(rt)) {
4150 resrtextents = qblocks = (uint)(e - s);
4151 resrtextents /= mp->m_sb.sb_rextsize;
4152 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4153 quota_flag = XFS_QMOPT_RES_RTBLKS;
4154 } else {
4155 resrtextents = 0;
4156 resblks = qblocks = \
4157 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4158 quota_flag = XFS_QMOPT_RES_REGBLKS;
4162 * Allocate and setup the transaction.
4164 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4165 error = xfs_trans_reserve(tp, resblks,
4166 XFS_WRITE_LOG_RES(mp), resrtextents,
4167 XFS_TRANS_PERM_LOG_RES,
4168 XFS_WRITE_LOG_COUNT);
4170 * Check for running out of space
4172 if (error) {
4174 * Free the transaction structure.
4176 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4177 xfs_trans_cancel(tp, 0);
4178 break;
4180 xfs_ilock(ip, XFS_ILOCK_EXCL);
4181 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4182 qblocks, 0, quota_flag);
4183 if (error)
4184 goto error1;
4186 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4187 xfs_trans_ihold(tp, ip);
4190 * Issue the xfs_bmapi() call to allocate the blocks
4192 XFS_BMAP_INIT(&free_list, &firstfsb);
4193 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4194 allocatesize_fsb, bmapi_flag,
4195 &firstfsb, 0, imapp, &nimaps,
4196 &free_list, NULL);
4197 if (error) {
4198 goto error0;
4202 * Complete the transaction
4204 error = xfs_bmap_finish(&tp, &free_list, &committed);
4205 if (error) {
4206 goto error0;
4209 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4210 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4211 if (error) {
4212 break;
4215 allocated_fsb = imapp->br_blockcount;
4217 if (nimaps == 0) {
4218 error = XFS_ERROR(ENOSPC);
4219 break;
4222 startoffset_fsb += allocated_fsb;
4223 allocatesize_fsb -= allocated_fsb;
4225 dmapi_enospc_check:
4226 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4227 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4229 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4230 XFS_ITOV(ip), DM_RIGHT_NULL,
4231 XFS_ITOV(ip), DM_RIGHT_NULL,
4232 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4233 if (error == 0)
4234 goto retry; /* Maybe DMAPI app. has made space */
4235 /* else fall through with error from XFS_SEND_DATA */
4238 return error;
4240 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4241 xfs_bmap_cancel(&free_list);
4242 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4244 error1: /* Just cancel transaction */
4245 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4246 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4247 goto dmapi_enospc_check;
4251 * Zero file bytes between startoff and endoff inclusive.
4252 * The iolock is held exclusive and no blocks are buffered.
4254 STATIC int
4255 xfs_zero_remaining_bytes(
4256 xfs_inode_t *ip,
4257 xfs_off_t startoff,
4258 xfs_off_t endoff)
4260 xfs_bmbt_irec_t imap;
4261 xfs_fileoff_t offset_fsb;
4262 xfs_off_t lastoffset;
4263 xfs_off_t offset;
4264 xfs_buf_t *bp;
4265 xfs_mount_t *mp = ip->i_mount;
4266 int nimap;
4267 int error = 0;
4269 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4270 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4271 mp->m_rtdev_targp : mp->m_ddev_targp);
4273 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4274 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4275 nimap = 1;
4276 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4277 NULL, 0, &imap, &nimap, NULL, NULL);
4278 if (error || nimap < 1)
4279 break;
4280 ASSERT(imap.br_blockcount >= 1);
4281 ASSERT(imap.br_startoff == offset_fsb);
4282 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4283 if (lastoffset > endoff)
4284 lastoffset = endoff;
4285 if (imap.br_startblock == HOLESTARTBLOCK)
4286 continue;
4287 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4288 if (imap.br_state == XFS_EXT_UNWRITTEN)
4289 continue;
4290 XFS_BUF_UNDONE(bp);
4291 XFS_BUF_UNWRITE(bp);
4292 XFS_BUF_READ(bp);
4293 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4294 xfsbdstrat(mp, bp);
4295 if ((error = xfs_iowait(bp))) {
4296 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4297 mp, bp, XFS_BUF_ADDR(bp));
4298 break;
4300 memset(XFS_BUF_PTR(bp) +
4301 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4302 0, lastoffset - offset + 1);
4303 XFS_BUF_UNDONE(bp);
4304 XFS_BUF_UNREAD(bp);
4305 XFS_BUF_WRITE(bp);
4306 xfsbdstrat(mp, bp);
4307 if ((error = xfs_iowait(bp))) {
4308 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4309 mp, bp, XFS_BUF_ADDR(bp));
4310 break;
4313 xfs_buf_free(bp);
4314 return error;
4318 * xfs_free_file_space()
4319 * This routine frees disk space for the given file.
4321 * This routine is only called by xfs_change_file_space
4322 * for an UNRESVSP type call.
4324 * RETURNS:
4325 * 0 on success
4326 * errno on error
4329 STATIC int
4330 xfs_free_file_space(
4331 xfs_inode_t *ip,
4332 xfs_off_t offset,
4333 xfs_off_t len,
4334 int attr_flags)
4336 bhv_vnode_t *vp;
4337 int committed;
4338 int done;
4339 xfs_off_t end_dmi_offset;
4340 xfs_fileoff_t endoffset_fsb;
4341 int error;
4342 xfs_fsblock_t firstfsb;
4343 xfs_bmap_free_t free_list;
4344 xfs_bmbt_irec_t imap;
4345 xfs_off_t ioffset;
4346 xfs_extlen_t mod=0;
4347 xfs_mount_t *mp;
4348 int nimap;
4349 uint resblks;
4350 uint rounding;
4351 int rt;
4352 xfs_fileoff_t startoffset_fsb;
4353 xfs_trans_t *tp;
4354 int need_iolock = 1;
4356 vp = XFS_ITOV(ip);
4357 mp = ip->i_mount;
4359 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4361 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4362 return error;
4364 error = 0;
4365 if (len <= 0) /* if nothing being freed */
4366 return error;
4367 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4368 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4369 end_dmi_offset = offset + len;
4370 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4372 if (offset < ip->i_size &&
4373 (attr_flags & ATTR_DMI) == 0 &&
4374 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4375 if (end_dmi_offset > ip->i_size)
4376 end_dmi_offset = ip->i_size;
4377 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4378 offset, end_dmi_offset - offset,
4379 AT_DELAY_FLAG(attr_flags), NULL);
4380 if (error)
4381 return error;
4384 if (attr_flags & ATTR_NOLOCK)
4385 need_iolock = 0;
4386 if (need_iolock) {
4387 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4388 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4391 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP);
4392 ioffset = offset & ~(rounding - 1);
4394 if (VN_CACHED(vp) != 0) {
4395 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4396 ctooff(offtoct(ioffset)), -1);
4397 error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
4398 -1, FI_REMAPF_LOCKED);
4399 if (error)
4400 goto out_unlock_iolock;
4404 * Need to zero the stuff we're not freeing, on disk.
4405 * If its a realtime file & can't use unwritten extents then we
4406 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4407 * will take care of it for us.
4409 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4410 nimap = 1;
4411 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4412 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4413 if (error)
4414 goto out_unlock_iolock;
4415 ASSERT(nimap == 0 || nimap == 1);
4416 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4417 xfs_daddr_t block;
4419 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4420 block = imap.br_startblock;
4421 mod = do_div(block, mp->m_sb.sb_rextsize);
4422 if (mod)
4423 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4425 nimap = 1;
4426 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4427 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4428 if (error)
4429 goto out_unlock_iolock;
4430 ASSERT(nimap == 0 || nimap == 1);
4431 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4432 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4433 mod++;
4434 if (mod && (mod != mp->m_sb.sb_rextsize))
4435 endoffset_fsb -= mod;
4438 if ((done = (endoffset_fsb <= startoffset_fsb)))
4440 * One contiguous piece to clear
4442 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4443 else {
4445 * Some full blocks, possibly two pieces to clear
4447 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4448 error = xfs_zero_remaining_bytes(ip, offset,
4449 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4450 if (!error &&
4451 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4452 error = xfs_zero_remaining_bytes(ip,
4453 XFS_FSB_TO_B(mp, endoffset_fsb),
4454 offset + len - 1);
4458 * free file space until done or until there is an error
4460 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4461 while (!error && !done) {
4464 * allocate and setup the transaction. Allow this
4465 * transaction to dip into the reserve blocks to ensure
4466 * the freeing of the space succeeds at ENOSPC.
4468 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4469 tp->t_flags |= XFS_TRANS_RESERVE;
4470 error = xfs_trans_reserve(tp,
4471 resblks,
4472 XFS_WRITE_LOG_RES(mp),
4474 XFS_TRANS_PERM_LOG_RES,
4475 XFS_WRITE_LOG_COUNT);
4478 * check for running out of space
4480 if (error) {
4482 * Free the transaction structure.
4484 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4485 xfs_trans_cancel(tp, 0);
4486 break;
4488 xfs_ilock(ip, XFS_ILOCK_EXCL);
4489 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4490 ip->i_udquot, ip->i_gdquot, resblks, 0,
4491 XFS_QMOPT_RES_REGBLKS);
4492 if (error)
4493 goto error1;
4495 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4496 xfs_trans_ihold(tp, ip);
4499 * issue the bunmapi() call to free the blocks
4501 XFS_BMAP_INIT(&free_list, &firstfsb);
4502 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4503 endoffset_fsb - startoffset_fsb,
4504 0, 2, &firstfsb, &free_list, NULL, &done);
4505 if (error) {
4506 goto error0;
4510 * complete the transaction
4512 error = xfs_bmap_finish(&tp, &free_list, &committed);
4513 if (error) {
4514 goto error0;
4517 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4518 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4521 out_unlock_iolock:
4522 if (need_iolock)
4523 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4524 return error;
4526 error0:
4527 xfs_bmap_cancel(&free_list);
4528 error1:
4529 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4530 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4531 XFS_ILOCK_EXCL);
4532 return error;
4536 * xfs_change_file_space()
4537 * This routine allocates or frees disk space for the given file.
4538 * The user specified parameters are checked for alignment and size
4539 * limitations.
4541 * RETURNS:
4542 * 0 on success
4543 * errno on error
4547 xfs_change_file_space(
4548 bhv_desc_t *bdp,
4549 int cmd,
4550 xfs_flock64_t *bf,
4551 xfs_off_t offset,
4552 cred_t *credp,
4553 int attr_flags)
4555 int clrprealloc;
4556 int error;
4557 xfs_fsize_t fsize;
4558 xfs_inode_t *ip;
4559 xfs_mount_t *mp;
4560 int setprealloc;
4561 xfs_off_t startoffset;
4562 xfs_off_t llen;
4563 xfs_trans_t *tp;
4564 bhv_vattr_t va;
4565 bhv_vnode_t *vp;
4567 vp = BHV_TO_VNODE(bdp);
4568 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4570 ip = XFS_BHVTOI(bdp);
4571 mp = ip->i_mount;
4574 * must be a regular file and have write permission
4576 if (!VN_ISREG(vp))
4577 return XFS_ERROR(EINVAL);
4579 xfs_ilock(ip, XFS_ILOCK_SHARED);
4581 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4582 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4583 return error;
4586 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4588 switch (bf->l_whence) {
4589 case 0: /*SEEK_SET*/
4590 break;
4591 case 1: /*SEEK_CUR*/
4592 bf->l_start += offset;
4593 break;
4594 case 2: /*SEEK_END*/
4595 bf->l_start += ip->i_size;
4596 break;
4597 default:
4598 return XFS_ERROR(EINVAL);
4601 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4603 if ( (bf->l_start < 0)
4604 || (bf->l_start > XFS_MAXIOFFSET(mp))
4605 || (bf->l_start + llen < 0)
4606 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4607 return XFS_ERROR(EINVAL);
4609 bf->l_whence = 0;
4611 startoffset = bf->l_start;
4612 fsize = ip->i_size;
4615 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4616 * file space.
4617 * These calls do NOT zero the data space allocated to the file,
4618 * nor do they change the file size.
4620 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4621 * space.
4622 * These calls cause the new file data to be zeroed and the file
4623 * size to be changed.
4625 setprealloc = clrprealloc = 0;
4627 switch (cmd) {
4628 case XFS_IOC_RESVSP:
4629 case XFS_IOC_RESVSP64:
4630 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4631 1, attr_flags);
4632 if (error)
4633 return error;
4634 setprealloc = 1;
4635 break;
4637 case XFS_IOC_UNRESVSP:
4638 case XFS_IOC_UNRESVSP64:
4639 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4640 attr_flags)))
4641 return error;
4642 break;
4644 case XFS_IOC_ALLOCSP:
4645 case XFS_IOC_ALLOCSP64:
4646 case XFS_IOC_FREESP:
4647 case XFS_IOC_FREESP64:
4648 if (startoffset > fsize) {
4649 error = xfs_alloc_file_space(ip, fsize,
4650 startoffset - fsize, 0, attr_flags);
4651 if (error)
4652 break;
4655 va.va_mask = XFS_AT_SIZE;
4656 va.va_size = startoffset;
4658 error = xfs_setattr(bdp, &va, attr_flags, credp);
4660 if (error)
4661 return error;
4663 clrprealloc = 1;
4664 break;
4666 default:
4667 ASSERT(0);
4668 return XFS_ERROR(EINVAL);
4672 * update the inode timestamp, mode, and prealloc flag bits
4674 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4676 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4677 0, 0, 0))) {
4678 /* ASSERT(0); */
4679 xfs_trans_cancel(tp, 0);
4680 return error;
4683 xfs_ilock(ip, XFS_ILOCK_EXCL);
4685 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4686 xfs_trans_ihold(tp, ip);
4688 if ((attr_flags & ATTR_DMI) == 0) {
4689 ip->i_d.di_mode &= ~S_ISUID;
4692 * Note that we don't have to worry about mandatory
4693 * file locking being disabled here because we only
4694 * clear the S_ISGID bit if the Group execute bit is
4695 * on, but if it was on then mandatory locking wouldn't
4696 * have been enabled.
4698 if (ip->i_d.di_mode & S_IXGRP)
4699 ip->i_d.di_mode &= ~S_ISGID;
4701 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4703 if (setprealloc)
4704 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4705 else if (clrprealloc)
4706 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4708 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4709 xfs_trans_set_sync(tp);
4711 error = xfs_trans_commit(tp, 0);
4713 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4715 return error;
4718 bhv_vnodeops_t xfs_vnodeops = {
4719 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4720 .vop_open = xfs_open,
4721 .vop_read = xfs_read,
4722 #ifdef HAVE_SPLICE
4723 .vop_splice_read = xfs_splice_read,
4724 .vop_splice_write = xfs_splice_write,
4725 #endif
4726 .vop_write = xfs_write,
4727 .vop_ioctl = xfs_ioctl,
4728 .vop_getattr = xfs_getattr,
4729 .vop_setattr = xfs_setattr,
4730 .vop_access = xfs_access,
4731 .vop_lookup = xfs_lookup,
4732 .vop_create = xfs_create,
4733 .vop_remove = xfs_remove,
4734 .vop_link = xfs_link,
4735 .vop_rename = xfs_rename,
4736 .vop_mkdir = xfs_mkdir,
4737 .vop_rmdir = xfs_rmdir,
4738 .vop_readdir = xfs_readdir,
4739 .vop_symlink = xfs_symlink,
4740 .vop_readlink = xfs_readlink,
4741 .vop_fsync = xfs_fsync,
4742 .vop_inactive = xfs_inactive,
4743 .vop_fid2 = xfs_fid2,
4744 .vop_rwlock = xfs_rwlock,
4745 .vop_rwunlock = xfs_rwunlock,
4746 .vop_bmap = xfs_bmap,
4747 .vop_reclaim = xfs_reclaim,
4748 .vop_attr_get = xfs_attr_get,
4749 .vop_attr_set = xfs_attr_set,
4750 .vop_attr_remove = xfs_attr_remove,
4751 .vop_attr_list = xfs_attr_list,
4752 .vop_link_removed = (vop_link_removed_t)fs_noval,
4753 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4754 .vop_tosspages = fs_tosspages,
4755 .vop_flushinval_pages = fs_flushinval_pages,
4756 .vop_flush_pages = fs_flush_pages,
4757 .vop_release = xfs_release,
4758 .vop_iflush = xfs_inode_flush,